[PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[René Bolldorf]

We don't need this ;-).

Best regards René Bolldorf & a happy new year in advance.

--- ./drivers/ata/libata-eh.c  2009-12-30 23:44:05.578988545 +0100
+++ ./drivers/ata/libata-eh.c  2009-12-30 23:45:06.991987607 +0100
@@ -1505,9 +1505,6 @@ static unsigned int atapi_eh_request_sen

  	DPRINTK("ATAPI request sense\n");

-	/* FIXME: is this needed? */
-	memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
-
  	/* initialize sense_buf with the error register,
  	 * for the case where they are -not- overwritten
  	 */

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[Jeff Garzik]

On 12/30/2009 05:59 PM, René Bolldorf wrote:
> We don't need this ;-).
>
> Best regards René Bolldorf & a happy new year in advance.
>
> --- ./drivers/ata/libata-eh.c 2009-12-30 23:44:05.578988545 +0100
> +++ ./drivers/ata/libata-eh.c 2009-12-30 23:45:06.991987607 +0100
> @@ -1505,9 +1505,6 @@ static unsigned int atapi_eh_request_sen
>
> DPRINTK("ATAPI request sense\n");
>
> - /* FIXME: is this needed? */
> - memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);

I need a little bit more detail than an unqualified statement...  Did 
you audit all paths leading to this code point?

	Jeff

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[René Bolldorf]

On 01/07/10 20:15, Jeff Garzik wrote:
> I need a little bit more detail than an unqualified statement...  Did
> you audit all paths leading to this code point?

Yes, and my two systems running fine with the patch, no oops or panic's.

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[René Bolldorf]

On 01/07/10 21:34, René Bolldorf wrote:
> On 01/07/10 20:15, Jeff Garzik wrote:
>> I need a little bit more detail than an unqualified statement... Did
>> you audit all paths leading to this code point?
>
> Yes, and my two systems running fine with the patch, no oops or panic's.

Sry forgot that:
	/* initialize sense_buf with the error register,
	 * for the case where they are -not- overwritten
	 */
	sense_buf[0] = 0x70;
	sense_buf[2] = dfl_sense_key;

So i think memset() is not needed and works very well without it.

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[Rolf Eike Beer]

[-- Attachment #1: Type: Text/Plain, Size: 623 bytes --]

René Bolldorf wrote:
> On 01/07/10 21:34, René Bolldorf wrote:
> > On 01/07/10 20:15, Jeff Garzik wrote:
> >> I need a little bit more detail than an unqualified statement... Did
> >> you audit all paths leading to this code point?
> >
> > Yes, and my two systems running fine with the patch, no oops or panic's.
> 
> Sry forgot that:
> 	/* initialize sense_buf with the error register,
> 	 * for the case where they are -not- overwritten
> 	 */
> 	sense_buf[0] = 0x70;
> 	sense_buf[2] = dfl_sense_key;
> 
> So i think memset() is not needed and works very well without it.

What happens to sense_buf[1]?

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[James Bottomley]

On Thu, 2010-01-07 at 14:15 -0500, Jeff Garzik wrote:
> On 12/30/2009 05:59 PM, René Bolldorf wrote:
> > We don't need this ;-).
> >
> > Best regards René Bolldorf & a happy new year in advance.
> >
> > --- ./drivers/ata/libata-eh.c 2009-12-30 23:44:05.578988545 +0100
> > +++ ./drivers/ata/libata-eh.c 2009-12-30 23:45:06.991987607 +0100
> > @@ -1505,9 +1505,6 @@ static unsigned int atapi_eh_request_sen
> >
> > DPRINTK("ATAPI request sense\n");
> >
> > - /* FIXME: is this needed? */
> > - memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
> 
> I need a little bit more detail than an unqualified statement...  Did 
> you audit all paths leading to this code point?

There are two code paths coming into here.  One directly from the scsi
sense buffer:

		if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
			tmp = atapi_eh_request_sense(qc->dev,
						qc->scsicmd->sense_buffer,
						qc->result_tf.feature >> 4);

Which is fine because SCSI zeros the sense buffer.

But one also here:

		u8 *sense_buffer = dev->link->ap->sector_buf;
		[...]
		err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);

Which doesn't look OK because it looks like the sector_buf isn't cleared
(and it is reused).

James

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[René Bolldorf]

On 01/08/10 16:28, James Bottomley wrote:
> On Thu, 2010-01-07 at 14:15 -0500, Jeff Garzik wrote:
>> On 12/30/2009 05:59 PM, René Bolldorf wrote:
>>> We don't need this ;-).
>>>
>>> Best regards René Bolldorf&  a happy new year in advance.
>>>
>>> --- ./drivers/ata/libata-eh.c 2009-12-30 23:44:05.578988545 +0100
>>> +++ ./drivers/ata/libata-eh.c 2009-12-30 23:45:06.991987607 +0100
>>> @@ -1505,9 +1505,6 @@ static unsigned int atapi_eh_request_sen
>>>
>>> DPRINTK("ATAPI request sense\n");
>>>
>>> - /* FIXME: is this needed? */
>>> - memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
>>
>> I need a little bit more detail than an unqualified statement...  Did
>> you audit all paths leading to this code point?
>
> There are two code paths coming into here.  One directly from the scsi
> sense buffer:
>
> 		if (!(qc->ap->pflags&  ATA_PFLAG_FROZEN)) {
> 			tmp = atapi_eh_request_sense(qc->dev,
> 						qc->scsicmd->sense_buffer,
> 						qc->result_tf.feature>>  4);
>
> Which is fine because SCSI zeros the sense buffer.
>
> But one also here:
>
> 		u8 *sense_buffer = dev->link->ap->sector_buf;
> 		[...]
> 		err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
>
> Which doesn't look OK because it looks like the sector_buf isn't cleared
> (and it is reused).
>
> James
>
>

Thank's, you're right. I have overseen this, sry for that.

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[Marc Bejarano]

2010/1/11 René Bolldorf <xsecute@googlemail.com>:
> I hope that's good enough explained :-).

definitely an improvement, but i think jeff is going to want your
S-O-B and at least a patch title, if no description :)


> +       /* make sure sense_buf is cleared then atapi_eh_request_sense is
> called.

s/then/when/ ?

> +        * (to make sure nothing get's reused.)

probably not necessary

> +        * thanks to James Bottomley

this may be better in a patch description

cheers,
marc

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[René Bolldorf]

On 01/11/10 20:41, Marc Bejarano wrote:
> and once more in plain text.  (sorry vger)
>
> 2010/1/11 Marc Bejarano<beej@beej.org>:
>> oops.  seems that GMANE's reply function only goes to a single "newsgroup".
>>   original recipients re-added.
>> marc
>> ----
>> From: Marc Bejarano<beej<at>  beej.org>
>> Subject: Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()
>> Newsgroups: gmane.linux.scsi
>> Date: 2010-01-08 22:21:09 GMT (2 days, 20 hours and 40 minutes ago)
>>
>> René Bolldorf<xsecute<at>  googlemail.com>  writes:
>>> On 01/08/10 16:28, James Bottomley wrote:
>>>> On Thu, 2010-01-07 at 14:15 -0500, Jeff Garzik wrote:
>>>>> On 12/30/2009 05:59 PM, René Bolldorf wrote:
>>>>>> We don't need this .
>>
>>>>>> - /* FIXME: is this needed? */
>>>>>> - memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
>>>>>
>>>>> I need a little bit more detail than an unqualified statement...  Did
>>>>> you audit all paths leading to this code point?
>>
>>>> But one also here:
>>>>
>>>> u8 *sense_buffer = dev->link->ap->sector_buf;
>>>> [...]
>>>> err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
>>>>
>>>> Which doesn't look OK because it looks like the sector_buf isn't cleared
>>>> (and it is reused).
>>>>
>>>> James
>>>>
>>>>
>>>
>>> Thank's, you're right. I have overseen this, sry for that.
>>
>> René: perhaps you'd like to submit a patch that substitutes FIXME comment
>> for one that explains why the memset is needed, crediting James in the
>> description?  we may as well gain something permanent from this discussion
>> that you started :)
>>
>> cheers,
>> marc
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
>> the body of a message to majordomo<at>  vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>> ----

I hope that's good enough explained :-).

Best regards
René

--- ./drivers/ata/libata-eh.c   2010-01-07 23:21:23.622464012 +0100
+++ ./drivers/ata/libata-eh.c   2010-01-11 21:11:19.869092897 +0100
@@ -1505,7 +1505,10 @@ static unsigned int atapi_eh_request_sen

         DPRINTK("ATAPI request sense\n");

-       /* FIXME: is this needed? */
+       /* make sure sense_buf is cleared then atapi_eh_request_sense is 
called.
+        * (to make sure nothing get's reused.)
+        * thanks to James Bottomley
+        */
         memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);

         /* initialize sense_buf with the error register,

Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()

[Marc Bejarano]

and once more in plain text.  (sorry vger)

2010/1/11 Marc Bejarano <beej@beej.org>:
> oops.  seems that GMANE's reply function only goes to a single "newsgroup".
>  original recipients re-added.
> marc
> ----
> From: Marc Bejarano <beej <at> beej.org>
> Subject: Re: [PATCH]drivers/ata/libata-eh.c:1509 unneeded memset()
> Newsgroups: gmane.linux.scsi
> Date: 2010-01-08 22:21:09 GMT (2 days, 20 hours and 40 minutes ago)
>
> René Bolldorf <xsecute <at> googlemail.com> writes:
>> On 01/08/10 16:28, James Bottomley wrote:
>> > On Thu, 2010-01-07 at 14:15 -0500, Jeff Garzik wrote:
>> >> On 12/30/2009 05:59 PM, René Bolldorf wrote:
>> >>> We don't need this .
>
>> >>> - /* FIXME: is this needed? */
>> >>> - memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
>> >>
>> >> I need a little bit more detail than an unqualified statement...  Did
>> >> you audit all paths leading to this code point?
>
>> > But one also here:
>> >
>> > u8 *sense_buffer = dev->link->ap->sector_buf;
>> > [...]
>> > err_mask = atapi_eh_request_sense(dev, sense_buffer, sense_key);
>> >
>> > Which doesn't look OK because it looks like the sector_buf isn't cleared
>> > (and it is reused).
>> >
>> > James
>> >
>> >
>>
>> Thank's, you're right. I have overseen this, sry for that.
>
> René: perhaps you'd like to submit a patch that substitutes FIXME comment
> for one that explains why the memset is needed, crediting James in the
> description?  we may as well gain something permanent from this discussion
> that you started :)
>
> cheers,
> marc
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
> the body of a message to majordomo <at> vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
> ----