[2/3] drivers/base: reorder consumer and its children behind suppliers
diff mbox series

Message ID 1529904187-18673-3-git-send-email-kernelfans@gmail.com
State Not Applicable
Headers show
Series
  • drivers/base: bugfix for supplier<-consumer ordering in device_kset
Related show

Commit Message

Pingfan Liu June 25, 2018, 5:23 a.m. UTC
commit 52cdbdd49853 ("driver core: correct device's shutdown order")
introduces supplier<-consumer order in devices_kset. The commit tries
to cleverly maintain both parent<-child and supplier<-consumer order by
reordering a device when probing. This method makes things simple and
clean, but unfortunately, breaks parent<-child order in some case,
which is described in next patch in this series.
Here this patch tries to resolve supplier<-consumer by only reordering a
device when it has suppliers, and takes care of the following scenario:
    [consumer, children] [ ... potential ... ] supplier
                         ^                   ^
After moving the consumer and its children after the supplier, the
potentail section may contain consumers whose supplier is inside
children, and this poses the requirement to dry out all consumpers in
the section recursively.

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Grygorii Strashko <grygorii.strashko@ti.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Bjorn Helgaas <helgaas@kernel.org>
Cc: Dave Young <dyoung@redhat.com>
Cc: linux-pci@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Pingfan Liu <kernelfans@gmail.com>
---
 drivers/base/base.h |  1 +
 drivers/base/core.c | 97 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 98 insertions(+)

Comments

Dan Carpenter June 26, 2018, 7:44 a.m. UTC | #1
[ There is a bug with kbuild where it's not showing the Smatch warnings
  but I can probably guess...  - dan ]

Hi Pingfan,

Thank you for the patch! Perhaps something to improve:

url:    https://github.com/0day-ci/linux/commits/Pingfan-Liu/drivers-base-bugfix-for-supplier-consumer-ordering-in-device_kset/20180625-132702


# https://github.com/0day-ci/linux/commit/1b2a1e63898baf80e8e830991284e1534bc54766
git remote add linux-review https://github.com/0day-ci/linux
git remote update linux-review
git checkout 1b2a1e63898baf80e8e830991284e1534bc54766
vim +/ret +245 drivers/base/core.c

1b2a1e63 Pingfan Liu 2018-06-25  216  
1b2a1e63 Pingfan Liu 2018-06-25  217  /* When reodering, take care of the range of (old_pos(dev), new_pos(dev)),
1b2a1e63 Pingfan Liu 2018-06-25  218   * there may be requirement to recursively move item.
1b2a1e63 Pingfan Liu 2018-06-25  219   */
1b2a1e63 Pingfan Liu 2018-06-25  220  int device_reorder_consumer(struct device *dev)
1b2a1e63 Pingfan Liu 2018-06-25  221  {
1b2a1e63 Pingfan Liu 2018-06-25  222  	struct list_head *iter, *left, *right;
1b2a1e63 Pingfan Liu 2018-06-25  223  	struct device *cur_dev;
1b2a1e63 Pingfan Liu 2018-06-25  224  	struct pos_info info;
1b2a1e63 Pingfan Liu 2018-06-25  225  	int ret, idx;
1b2a1e63 Pingfan Liu 2018-06-25  226  
1b2a1e63 Pingfan Liu 2018-06-25  227  	idx = device_links_read_lock();
1b2a1e63 Pingfan Liu 2018-06-25  228  	if (list_empty(&dev->links.suppliers)) {
1b2a1e63 Pingfan Liu 2018-06-25  229  		device_links_read_unlock(idx);
1b2a1e63 Pingfan Liu 2018-06-25  230  		return 0;
1b2a1e63 Pingfan Liu 2018-06-25  231  	}
1b2a1e63 Pingfan Liu 2018-06-25  232  	spin_lock(&devices_kset->list_lock);
1b2a1e63 Pingfan Liu 2018-06-25  233  	list_for_each_prev(iter, &devices_kset->list) {
1b2a1e63 Pingfan Liu 2018-06-25  234  		cur_dev = list_entry(iter, struct device, kobj.entry);
1b2a1e63 Pingfan Liu 2018-06-25  235  		ret = find_last_supplier(dev, cur_dev);
1b2a1e63 Pingfan Liu 2018-06-25  236  		switch (ret) {
1b2a1e63 Pingfan Liu 2018-06-25  237  		case -1:
1b2a1e63 Pingfan Liu 2018-06-25  238  			goto unlock;
1b2a1e63 Pingfan Liu 2018-06-25  239  		case 1:
1b2a1e63 Pingfan Liu 2018-06-25  240  			break;
1b2a1e63 Pingfan Liu 2018-06-25  241  		case 0:
1b2a1e63 Pingfan Liu 2018-06-25  242  			continue;

The break breaks from the switch and the continue continues the loop so
they're equivalent.  Perhaps you intended to break from the loop?

1b2a1e63 Pingfan Liu 2018-06-25  243  		}
1b2a1e63 Pingfan Liu 2018-06-25  244  	}
1b2a1e63 Pingfan Liu 2018-06-25 @245  	BUG_ON(!ret);

If the list is empty then "ret" can be unitialized.  We test a different
list "dev->links.suppliers" to see if that's empty.  I wrote a bunch of
code to make Smatch try to understand about empty lists, but I don't
think it works...

1b2a1e63 Pingfan Liu 2018-06-25  246  
1b2a1e63 Pingfan Liu 2018-06-25  247  	/* record the affected open section */
1b2a1e63 Pingfan Liu 2018-06-25  248  	left = dev->kobj.entry.prev;
1b2a1e63 Pingfan Liu 2018-06-25  249  	right = iter;
1b2a1e63 Pingfan Liu 2018-06-25  250  	info.pos = list_entry(iter, struct device, kobj.entry);
1b2a1e63 Pingfan Liu 2018-06-25  251  	info.tail = NULL;
1b2a1e63 Pingfan Liu 2018-06-25  252  	/* dry out the consumers in (left,right) */
1b2a1e63 Pingfan Liu 2018-06-25  253  	__device_reorder_consumer(dev, left, right, &info);
1b2a1e63 Pingfan Liu 2018-06-25  254  
1b2a1e63 Pingfan Liu 2018-06-25  255  unlock:
1b2a1e63 Pingfan Liu 2018-06-25  256  	spin_unlock(&devices_kset->list_lock);
1b2a1e63 Pingfan Liu 2018-06-25  257  	device_links_read_unlock(idx);
1b2a1e63 Pingfan Liu 2018-06-25  258  	return 0;
1b2a1e63 Pingfan Liu 2018-06-25  259  }
1b2a1e63 Pingfan Liu 2018-06-25  260  

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
Pingfan Liu June 27, 2018, 2:34 a.m. UTC | #2
Hi Dan,

Thanks for your hints, see the comment in lines.

On Tue, Jun 26, 2018 at 3:44 PM Dan Carpenter <dan.carpenter@oracle.com> wrote:
>
> [ There is a bug with kbuild where it's not showing the Smatch warnings
>   but I can probably guess...  - dan ]
>
> Hi Pingfan,
>
> Thank you for the patch! Perhaps something to improve:
>
> url:    https://github.com/0day-ci/linux/commits/Pingfan-Liu/drivers-base-bugfix-for-supplier-consumer-ordering-in-device_kset/20180625-132702
>
>
> # https://github.com/0day-ci/linux/commit/1b2a1e63898baf80e8e830991284e1534bc54766
> git remote add linux-review https://github.com/0day-ci/linux
> git remote update linux-review
> git checkout 1b2a1e63898baf80e8e830991284e1534bc54766
> vim +/ret +245 drivers/base/core.c
>
> 1b2a1e63 Pingfan Liu 2018-06-25  216
> 1b2a1e63 Pingfan Liu 2018-06-25  217  /* When reodering, take care of the range of (old_pos(dev), new_pos(dev)),
> 1b2a1e63 Pingfan Liu 2018-06-25  218   * there may be requirement to recursively move item.
> 1b2a1e63 Pingfan Liu 2018-06-25  219   */
> 1b2a1e63 Pingfan Liu 2018-06-25  220  int device_reorder_consumer(struct device *dev)
> 1b2a1e63 Pingfan Liu 2018-06-25  221  {
> 1b2a1e63 Pingfan Liu 2018-06-25  222    struct list_head *iter, *left, *right;
> 1b2a1e63 Pingfan Liu 2018-06-25  223    struct device *cur_dev;
> 1b2a1e63 Pingfan Liu 2018-06-25  224    struct pos_info info;
> 1b2a1e63 Pingfan Liu 2018-06-25  225    int ret, idx;
> 1b2a1e63 Pingfan Liu 2018-06-25  226
> 1b2a1e63 Pingfan Liu 2018-06-25  227    idx = device_links_read_lock();
> 1b2a1e63 Pingfan Liu 2018-06-25  228    if (list_empty(&dev->links.suppliers)) {
> 1b2a1e63 Pingfan Liu 2018-06-25  229            device_links_read_unlock(idx);
> 1b2a1e63 Pingfan Liu 2018-06-25  230            return 0;
> 1b2a1e63 Pingfan Liu 2018-06-25  231    }
> 1b2a1e63 Pingfan Liu 2018-06-25  232    spin_lock(&devices_kset->list_lock);
> 1b2a1e63 Pingfan Liu 2018-06-25  233    list_for_each_prev(iter, &devices_kset->list) {
> 1b2a1e63 Pingfan Liu 2018-06-25  234            cur_dev = list_entry(iter, struct device, kobj.entry);
> 1b2a1e63 Pingfan Liu 2018-06-25  235            ret = find_last_supplier(dev, cur_dev);
> 1b2a1e63 Pingfan Liu 2018-06-25  236            switch (ret) {
> 1b2a1e63 Pingfan Liu 2018-06-25  237            case -1:
> 1b2a1e63 Pingfan Liu 2018-06-25  238                    goto unlock;
> 1b2a1e63 Pingfan Liu 2018-06-25  239            case 1:
> 1b2a1e63 Pingfan Liu 2018-06-25  240                    break;
> 1b2a1e63 Pingfan Liu 2018-06-25  241            case 0:
> 1b2a1e63 Pingfan Liu 2018-06-25  242                    continue;
>
> The break breaks from the switch and the continue continues the loop so
> they're equivalent.  Perhaps you intended to break from the loop?
>
Yes, you are right.

> 1b2a1e63 Pingfan Liu 2018-06-25  243            }
> 1b2a1e63 Pingfan Liu 2018-06-25  244    }
> 1b2a1e63 Pingfan Liu 2018-06-25 @245    BUG_ON(!ret);
>
> If the list is empty then "ret" can be unitialized.  We test a different
> list "dev->links.suppliers" to see if that's empty.  I wrote a bunch of
> code to make Smatch try to understand about empty lists, but I don't
> think it works...
>
Yes, if list_empty, then the code can not touch ret. But ret is
useless in this scene. Does it matter?

Thanks and regards,
Pingfan

> 1b2a1e63 Pingfan Liu 2018-06-25  246
> 1b2a1e63 Pingfan Liu 2018-06-25  247    /* record the affected open section */
> 1b2a1e63 Pingfan Liu 2018-06-25  248    left = dev->kobj.entry.prev;
> 1b2a1e63 Pingfan Liu 2018-06-25  249    right = iter;
> 1b2a1e63 Pingfan Liu 2018-06-25  250    info.pos = list_entry(iter, struct device, kobj.entry);
> 1b2a1e63 Pingfan Liu 2018-06-25  251    info.tail = NULL;
> 1b2a1e63 Pingfan Liu 2018-06-25  252    /* dry out the consumers in (left,right) */
> 1b2a1e63 Pingfan Liu 2018-06-25  253    __device_reorder_consumer(dev, left, right, &info);
> 1b2a1e63 Pingfan Liu 2018-06-25  254
> 1b2a1e63 Pingfan Liu 2018-06-25  255  unlock:
> 1b2a1e63 Pingfan Liu 2018-06-25  256    spin_unlock(&devices_kset->list_lock);
> 1b2a1e63 Pingfan Liu 2018-06-25  257    device_links_read_unlock(idx);
> 1b2a1e63 Pingfan Liu 2018-06-25  258    return 0;
> 1b2a1e63 Pingfan Liu 2018-06-25  259  }
> 1b2a1e63 Pingfan Liu 2018-06-25  260
>
> ---
> 0-DAY kernel test infrastructure                Open Source Technology Center
> https://lists.01.org/pipermail/kbuild-all                   Intel Corporation
Dan Carpenter June 27, 2018, 8:34 a.m. UTC | #3
On Wed, Jun 27, 2018 at 10:34:54AM +0800, Pingfan Liu wrote:
> > 1b2a1e63 Pingfan Liu 2018-06-25  243            }
> > 1b2a1e63 Pingfan Liu 2018-06-25  244    }
> > 1b2a1e63 Pingfan Liu 2018-06-25 @245    BUG_ON(!ret);
> >
> > If the list is empty then "ret" can be unitialized.  We test a different
> > list "dev->links.suppliers" to see if that's empty.  I wrote a bunch of
> > code to make Smatch try to understand about empty lists, but I don't
> > think it works...
> >
> Yes, if list_empty, then the code can not touch ret. But ret is
> useless in this scene. Does it matter?
> 

I'm not sure I understand what you're asking?  Of course, it matters?

regards,
dan carpenter
Pingfan Liu June 28, 2018, 1:47 p.m. UTC | #4
On Wed, Jun 27, 2018 at 4:35 PM Dan Carpenter <dan.carpenter@oracle.com> wrote:
>
> On Wed, Jun 27, 2018 at 10:34:54AM +0800, Pingfan Liu wrote:
> > > 1b2a1e63 Pingfan Liu 2018-06-25  243            }
> > > 1b2a1e63 Pingfan Liu 2018-06-25  244    }
> > > 1b2a1e63 Pingfan Liu 2018-06-25 @245    BUG_ON(!ret);
> > >
> > > If the list is empty then "ret" can be unitialized.  We test a different
> > > list "dev->links.suppliers" to see if that's empty.  I wrote a bunch of
> > > code to make Smatch try to understand about empty lists, but I don't
> > > think it works...
> > >
> > Yes, if list_empty, then the code can not touch ret. But ret is
> > useless in this scene. Does it matter?
> >
>
> I'm not sure I understand what you're asking?  Of course, it matters?
>
Oh, I misunderstood your original comment. Yes, you are right. I will
fix it in next version, if this code section is still used.

Thanks and regards,
Pingfan

Patch
diff mbox series

diff --git a/drivers/base/base.h b/drivers/base/base.h
index a75c302..37f86ca 100644
--- a/drivers/base/base.h
+++ b/drivers/base/base.h
@@ -135,6 +135,7 @@  extern void device_unblock_probing(void);
 
 /* /sys/devices directory */
 extern struct kset *devices_kset;
+extern int device_reorder_consumer(struct device *dev);
 extern void devices_kset_move_last(struct device *dev);
 
 #if defined(CONFIG_MODULES) && defined(CONFIG_SYSFS)
diff --git a/drivers/base/core.c b/drivers/base/core.c
index 8113d2c..db30e86 100644
--- a/drivers/base/core.c
+++ b/drivers/base/core.c
@@ -161,6 +161,103 @@  static bool is_consumer(struct device *query, struct device *supplier)
 	return false;
 }
 
+/* recursively move the potential consumers in open section (left, right)
+ * after the barrier
+ */
+static int __device_reorder_consumer(struct device *consumer,
+	struct list_head *left, struct list_head *right,
+	struct pos_info *p)
+{
+	struct list_head *iter;
+	struct device *c_dev, *s_dev, *tail_dev;
+
+	descendants_reorder_after_pos(consumer, p);
+	tail_dev = p->tail;
+	/* (left, right) may contain consumers, hence checking if any moved
+	 * child serving as supplier. The reversing order help us to meet
+	 * the last supplier of a consumer.
+	 */
+	list_opensect_for_each_reverse(iter, left, right) {
+		struct list_head *l_iter, *moved_left, *moved_right;
+
+		moved_left = (&consumer->kobj.entry)->prev;
+		moved_right = tail_dev->kobj.entry.next;
+		/* the moved section may contain potential suppliers */
+		list_opensect_for_each_reverse(l_iter, moved_left,
+			moved_right) {
+			s_dev = list_entry(l_iter, struct device, kobj.entry);
+			c_dev = list_entry(iter, struct device, kobj.entry);
+			/* to fix: this poses extra effort for locking */
+			if (is_consumer(c_dev, s_dev)) {
+				p->tail = NULL;
+				/* to fix: lock issue */
+				p->pos =  s_dev;
+				/* reorder after the last supplier */
+				__device_reorder_consumer(c_dev,
+					l_iter, right, p);
+			}
+		}
+	}
+	return 0;
+}
+
+static int find_last_supplier(struct device *dev, struct device *supplier)
+{
+	struct device_link *link;
+
+	list_for_each_entry_reverse(link, &dev->links.suppliers, c_node) {
+		if (link->supplier == supplier)
+			return 1;
+	}
+	if (dev == supplier)
+		return -1;
+	return 0;
+}
+
+/* When reodering, take care of the range of (old_pos(dev), new_pos(dev)),
+ * there may be requirement to recursively move item.
+ */
+int device_reorder_consumer(struct device *dev)
+{
+	struct list_head *iter, *left, *right;
+	struct device *cur_dev;
+	struct pos_info info;
+	int ret, idx;
+
+	idx = device_links_read_lock();
+	if (list_empty(&dev->links.suppliers)) {
+		device_links_read_unlock(idx);
+		return 0;
+	}
+	spin_lock(&devices_kset->list_lock);
+	list_for_each_prev(iter, &devices_kset->list) {
+		cur_dev = list_entry(iter, struct device, kobj.entry);
+		ret = find_last_supplier(dev, cur_dev);
+		switch (ret) {
+		case -1:
+			goto unlock;
+		case 1:
+			break;
+		case 0:
+			continue;
+		}
+	}
+	BUG_ON(!ret);
+
+	/* record the affected open section */
+	left = dev->kobj.entry.prev;
+	right = iter;
+	info.pos = list_entry(iter, struct device, kobj.entry);
+	info.tail = NULL;
+	/* dry out the consumers in (left,right) */
+	__device_reorder_consumer(dev, left, right, &info);
+
+unlock:
+	spin_unlock(&devices_kset->list_lock);
+	device_links_read_unlock(idx);
+	return 0;
+}
+
 static int device_reorder_to_tail(struct device *dev, void *not_used)
 {
 	struct device_link *link;