[PATCH] qlogicpti: Add missing parentheses

[PATCH] qlogicpti: Add missing parentheses

[Roel Kluin]

Add missing parentheses

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
---
This was used at drivers/scsi/qlogicpti.c:

946:   host->sg_tablesize = QLOGICPTI_MAX_SG(num_free);
1270:  .sg_tablesize           = QLOGICPTI_MAX_SG(QLOGICPTI_REQ_QUEUE_LEN),

diff --git a/drivers/scsi/qlogicpti.h b/drivers/scsi/qlogicpti.h
index 9c053bb..e3c74d1 100644
--- a/drivers/scsi/qlogicpti.h
+++ b/drivers/scsi/qlogicpti.h
@@ -43,7 +43,7 @@
  * determined for each queue request anew.
  */
 #define QLOGICPTI_REQ_QUEUE_LEN	255	/* must be power of two - 1 */
-#define QLOGICPTI_MAX_SG(ql)	(4 + ((ql) > 0) ? 7*((ql) - 1) : 0)
+#define QLOGICPTI_MAX_SG(ql)	(4 + (((ql) > 0) ? 7*((ql) - 1) : 0))
 
 /* mailbox command complete status codes */
 #define MBOX_COMMAND_COMPLETE		0x4000

Re: [PATCH] qlogicpti: Add missing parentheses

[David Miller]

From: Roel Kluin <roel.kluin@gmail.com>
Date: Wed, 18 Feb 2009 13:48:11 +0100

> Add missing parentheses
> 
> Signed-off-by: Roel Kluin <roel.kluin@gmail.com>

So what bug does this cause?

>  #define QLOGICPTI_REQ_QUEUE_LEN	255	/* must be power of two - 1 */
> -#define QLOGICPTI_MAX_SG(ql)	(4 + ((ql) > 0) ? 7*((ql) - 1) : 0)
> +#define QLOGICPTI_MAX_SG(ql)	(4 + (((ql) > 0) ? 7*((ql) - 1) : 0))

The "ql > 0" bids to the "?" operator, so this expression
does the right thing as-is as far as I can tell.

Do you think that "4 + ((ql > 0)" binds to "?", I don't think
it does.

Re: [PATCH] qlogicpti: Add missing parentheses

[Andrew Morton]

On Wed, 18 Feb 2009 14:39:16 -0800 (PST)
David Miller <davem@davemloft.net> wrote:

> From: Roel Kluin <roel.kluin@gmail.com>
> Date: Wed, 18 Feb 2009 13:48:11 +0100
> 
> > Add missing parentheses
> > 
> > Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
> 
> So what bug does this cause?
> 
> >  #define QLOGICPTI_REQ_QUEUE_LEN	255	/* must be power of two - 1 */
> > -#define QLOGICPTI_MAX_SG(ql)	(4 + ((ql) > 0) ? 7*((ql) - 1) : 0)
> > +#define QLOGICPTI_MAX_SG(ql)	(4 + (((ql) > 0) ? 7*((ql) - 1) : 0))
> 
> The "ql > 0" bids to the "?" operator, so this expression
> does the right thing as-is as far as I can tell.
> 
> Do you think that "4 + ((ql > 0)" binds to "?", I don't think
> it does.

The code is crap, and buggy if passed an expression with side-effects. 
Let's turn it into the way it _should_ have been written.

yes, "+" has higher precedence than ?:, so the current code is:

static inline int qlogicpti_max_sg(int ql)
{
	if (4 + (ql > 0))
		return 7 * (ql - 1);
	else
		return 0;
}

Now, 4 plus a boolean can only ever evaluate to 4 or 5, so this
function can never return zero.  So yeah, I assume that this was meant:

static inline int qlogicpti_max_sg(int ql)
{
	if (ql > 0)
		return 7 * (ql - 1) + 4;
	else
		return 4;
}

Re: [PATCH] qlogicpti: Add missing parentheses

[David Miller]

From: Andrew Morton <akpm@linux-foundation.org>
Date: Thu, 19 Feb 2009 13:01:42 -0800

> On Wed, 18 Feb 2009 14:39:16 -0800 (PST)
> David Miller <davem@davemloft.net> wrote:
> 
> > From: Roel Kluin <roel.kluin@gmail.com>
> > Date: Wed, 18 Feb 2009 13:48:11 +0100
> > 
> > > Add missing parentheses
> > > 
> > > Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
> > 
> > So what bug does this cause?
> > 
> > >  #define QLOGICPTI_REQ_QUEUE_LEN	255	/* must be power of two - 1 */
> > > -#define QLOGICPTI_MAX_SG(ql)	(4 + ((ql) > 0) ? 7*((ql) - 1) : 0)
> > > +#define QLOGICPTI_MAX_SG(ql)	(4 + (((ql) > 0) ? 7*((ql) - 1) : 0))
> > 
> > The "ql > 0" bids to the "?" operator, so this expression
> > does the right thing as-is as far as I can tell.
> > 
> > Do you think that "4 + ((ql > 0)" binds to "?", I don't think
> > it does.
> 
> The code is crap, and buggy if passed an expression with side-effects. 

Thank got it never is called that way :-)

> Now, 4 plus a boolean can only ever evaluate to 4 or 5, so this
> function can never return zero.  So yeah, I assume that this was meant:
> 
> static inline int qlogicpti_max_sg(int ql)
> {
> 	if (ql > 0)
> 		return 7 * (ql - 1) + 4;
> 	else
> 		return 4;
> }
> 

I think this doesn't even match the intention, and I wrote
the code so...

We want to return 0 if the queue has no free entries.
But if it does have free slots, we want "4 + xxx"

These problems probably explain the mysterious workaround
in update_can_queue().

[PATCH] qlogicpti: add missing parentheses

[Roel Kluin]

`+' has a higher precedence than `?' so the condition always
evaluates to true and this is preprocessed to `7*((ql) - 1)'

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
---
 drivers/scsi/qlogicpti.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

This one must be right, I think.

QLOGICPTI_MAX_SG is used here:

drivers/scsi/qlogicpti.c:964:   host->sg_tablesize = QLOGICPTI_MAX_SG(num_free);
drivers/scsi/qlogicpti.c:1288:  .sg_tablesize           = QLOGICPTI_MAX_SG(QLOGICPTI_REQ_QUEUE_LEN),

Thanks, Roel

diff --git a/drivers/scsi/qlogicpti.h b/drivers/scsi/qlogicpti.h
index 9c053bb..e3c74d1 100644
--- a/drivers/scsi/qlogicpti.h
+++ b/drivers/scsi/qlogicpti.h
@@ -43,7 +43,7 @@
  * determined for each queue request anew.
  */
 #define QLOGICPTI_REQ_QUEUE_LEN	255	/* must be power of two - 1 */
-#define QLOGICPTI_MAX_SG(ql)	(4 + ((ql) > 0) ? 7*((ql) - 1) : 0)
+#define QLOGICPTI_MAX_SG(ql)	(4 + (((ql) > 0) ? 7*((ql) - 1) : 0))
 
 /* mailbox command complete status codes */
 #define MBOX_COMMAND_COMPLETE		0x4000

Re: [PATCH] qlogicpti: add missing parentheses

[James Bottomley]

On Wed, 2009-11-04 at 00:38 +0100, Roel Kluin wrote:
> `+' has a higher precedence than `?' so the condition always
> evaluates to true and this is preprocessed to `7*((ql) - 1)'
> 
> Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
> ---
>  drivers/scsi/qlogicpti.h |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> This one must be right, I think.

It looks plausible ... but it will increase the mailbox segment length
by 4 ... someone needs to verify that this increase will actually work
and won't run off the end of the mailbox table.

James

Re: [PATCH] qlogicpti: add missing parentheses

[David Miller]

From: James Bottomley <James.Bottomley@suse.de>
Date: Tue, 03 Nov 2009 17:51:49 -0600

> On Wed, 2009-11-04 at 00:38 +0100, Roel Kluin wrote:
>> `+' has a higher precedence than `?' so the condition always
>> evaluates to true and this is preprocessed to `7*((ql) - 1)'
>> 
>> Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
>> ---
>>  drivers/scsi/qlogicpti.h |    2 +-
>>  1 files changed, 1 insertions(+), 1 deletions(-)
>> 
>> This one must be right, I think.
> 
> It looks plausible ... but it will increase the mailbox segment length
> by 4 ... someone needs to verify that this increase will actually work
> and won't run off the end of the mailbox table.

It should be fine, good find Roel.

Acked-by: David S. Miller <davem@davemloft.net>