Patchwork powerpc: hv{cs, si}_close() both unsigned hp->count and hvcsd->open_count cannot be negative

login
register
mail settings
Submitter roel kluin
Date Nov. 29, 2008, 10:57 a.m.
Message ID <4931201D.7060701@gmail.com>
Download mbox | patch
Permalink /patch/11394/
State Superseded
Headers show

Comments

roel kluin - Nov. 29, 2008, 10:57 a.m.
unsigned hp->count and hvcsd->open_count cannot be negative

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
---
Both members of respectively struct hvcs_struct, see
vi drivers/char/hvcs.c +262

and struct hvcs_struct, see
vi drivers/char/hvsi.c +70
Andreas Schwab - Nov. 29, 2008, 12:54 p.m.
roel kluin <roel.kluin@gmail.com> writes:

> -	if (--hvcsd->open_count == 0) {
> +	if (hvcsd->open_count == 1) {
> +		hvcsd->open_count--;

This is not the same.

> -	if (--hp->count == 0) {
> +	if (hp->count == 1) {
> +		hp->count--;

Likewise.

Andreas.
roel kluin - Nov. 29, 2008, 1:27 p.m.
Andreas Schwab wrote:
> roel kluin <roel.kluin@gmail.com> writes:
> 
>> -	if (--hvcsd->open_count == 0) {
>> +	if (hvcsd->open_count == 1) {
>> +		hvcsd->open_count--;
> 
> This is not the same.

I think you're missing that I also decrement if (hvcsd->open_count > 1)
If not, please elaborate.

> 
>> -	if (--hp->count == 0) {
>> +	if (hp->count == 1) {
>> +		hp->count--;
> 
> Likewise.

Same here.

Roel
Andreas Schwab - Nov. 29, 2008, 3:01 p.m.
roel kluin <roel.kluin@gmail.com> writes:

> Andreas Schwab wrote:
>> roel kluin <roel.kluin@gmail.com> writes:
>> 
>>> -	if (--hvcsd->open_count == 0) {
>>> +	if (hvcsd->open_count == 1) {
>>> +		hvcsd->open_count--;
>> 
>> This is not the same.
>
> I think you're missing that I also decrement if (hvcsd->open_count > 1)

Yes, sorry, I missed that.

Andreas.
Paul Mackerras - Nov. 29, 2008, 11:20 p.m.
roel kluin writes:

> unsigned hp->count and hvcsd->open_count cannot be negative
...
> -	if (--hvcsd->open_count == 0) {
> +	if (hvcsd->open_count == 1) {
> +		hvcsd->open_count--;

Why are we no longer decrementing hvcsd->open_count in the cases where
it is greater than 1?

> -	if (--hp->count == 0) {
> +	if (hp->count == 1) {
> +		hp->count--;

Ditto with hp->count here?

Also, I don't see why those changes have anything to do with "unsigned
things cannot be negative".  As long as those counts are never zero on
entry to those code sections, the existing code is fine, and I believe
that assertion can be maintained.  If you believe the code needs to
defend against the possibility of a zero count on entry, that should
have been explicitly stated in the patch description.

Paul.
Alan Cox - Nov. 30, 2008, 11:16 a.m.
> Also, I don't see why those changes have anything to do with "unsigned
> things cannot be negative".  As long as those counts are never zero on
> entry to those code sections, the existing code is fine, and I believe
> that assertion can be maintained.  If you believe the code needs to
> defend against the possibility of a zero count on entry, that should
> have been explicitly stated in the patch description.

See the tty_port patches in -next - this is an argument about code which
ought eventually to go away replaced by standard helper functions.
Paul Mackerras - Nov. 30, 2008, 9:23 p.m.
Alan Cox writes:

> See the tty_port patches in -next - this is an argument about code which
> ought eventually to go away replaced by standard helper functions.

OK.  In that case I think the best fix for the existing code is just
to change the count to be int rather than unsigned int.

Paul.

Patch

diff --git a/drivers/char/hvcs.c b/drivers/char/hvcs.c
index 473d9b1..b228b84 100644
--- a/drivers/char/hvcs.c
+++ b/drivers/char/hvcs.c
@@ -1222,7 +1222,8 @@  static void hvcs_close(struct tty_struct *tty, struct file *filp)
 	hvcsd = tty->driver_data;
 
 	spin_lock_irqsave(&hvcsd->lock, flags);
-	if (--hvcsd->open_count == 0) {
+	if (hvcsd->open_count == 1) {
+		hvcsd->open_count--;
 
 		vio_disable_interrupts(hvcsd->vdev);
 
@@ -1248,7 +1249,9 @@  static void hvcs_close(struct tty_struct *tty, struct file *filp)
 		free_irq(irq, hvcsd);
 		kref_put(&hvcsd->kref, destroy_hvcs_struct);
 		return;
-	} else if (hvcsd->open_count < 0) {
+	} else if (hvcsd->open_count > 1) {
+		hvcsd->open_count--;
+	} else {
 		printk(KERN_ERR "HVCS: vty-server@%X open_count: %d"
 				" is missmanaged.\n",
 		hvcsd->vdev->unit_address, hvcsd->open_count);
diff --git a/drivers/char/hvsi.c b/drivers/char/hvsi.c
index 59c6f9a..d46bccd 100644
--- a/drivers/char/hvsi.c
+++ b/drivers/char/hvsi.c
@@ -875,7 +875,8 @@  static void hvsi_close(struct tty_struct *tty, struct file *filp)
 
 	spin_lock_irqsave(&hp->lock, flags);
 
-	if (--hp->count == 0) {
+	if (hp->count == 1) {
+		hp->count--;
 		hp->tty = NULL;
 		hp->inbuf_end = hp->inbuf; /* discard remaining partial packets */
 
@@ -908,7 +909,9 @@  static void hvsi_close(struct tty_struct *tty, struct file *filp)
 
 			spin_lock_irqsave(&hp->lock, flags);
 		}
-	} else if (hp->count < 0)
+	} else if (hp->count > 1)
+		hp->count--;
+	else
 		printk(KERN_ERR "hvsi_close %lu: oops, count is %d\n",
 		       hp - hvsi_ports, hp->count);