From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mlbe2k1.cs.myharris.net (mlbe2k1.cs.myharris.net [137.237.90.88]) by ozlabs.org (Postfix) with ESMTP id 4F754DE4E5 for ; Tue, 12 Aug 2008 02:42:49 +1000 (EST) Message-ID: <48A06C06.2070503@harris.com> Date: Mon, 11 Aug 2008 12:42:46 -0400 From: "Steven A. Falco" MIME-Version: 1.0 To: Ben Dooks Subject: Re: [PATCH] pata_of_platform: fix no irq handling References: <48A05152.7020508@harris.com> <20080811151913.GA14690@oksana.dev.rtsoft.ru> <20080811163648.GI26082@trinity.fluff.org> In-Reply-To: <20080811163648.GI26082@trinity.fluff.org> Content-Type: multipart/alternative; boundary="------------070500030100010105040908" Cc: linuxppc-dev@ozlabs.org, Jeff Garzik , linux-ide@vger.kernel.org List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , This is a multi-part message in MIME format. --------------070500030100010105040908 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Ben Dooks wrote: > On Mon, Aug 11, 2008 at 07:19:13PM +0400, Anton Vorontsov wrote: > >> When no irq specified, pata_of_platform fills irq_res with -1, >> which is wrong to do for two reasons: >> >> 1. By definition, 'no irq' should be IRQ 0, not some negative integer; >> > > interesting, IRQ 0 is actually valid on some ARM systems. > > It is here too, but I believe most of the code uses a virtualized irq number, so physical IRQ 0 would presumably get mapped to a non-zero virtual one. >> 2. pata_platform checks for irq_res.start > 0, but since irq_res.start >> is unsigned type, the check will be true for `-1'. >> >> Reported-by: Steven A. Falco >> Signed-off-by: Anton Vorontsov >> --- >> >> On Mon, Aug 11, 2008 at 10:48:50AM -0400, Steven A. Falco wrote: >> >>> I think there is a bug in the communications between pata_of_platform >>> and pata_platform. I will refer to the master branch of the DENX git >>> tree, which is roughly v2.6.26.1 at this time. I am using a Sequoia >>> board with a PPC440EPx. >>> >>> In pata_of_platform, we have: >>> >>> ret = of_irq_to_resource(dn, 0, &irq_res); >>> if (ret == NO_IRQ) >>> irq_res.start = irq_res.end = -1; >>> >>> so if there is no interrupt defined, then start and end are -1. >>> However, __pata_platform_probe has: >>> >>> if (irq_res && irq_res->start > 0) { >>> irq = irq_res->start; >>> irq_flags = irq_res->flags; >>> } >>> >>> You might think that the (irq_res->start > 0) test will fail, as it >>> should in this no-irq case. But, start is a u64, so the -1 actually >>> looks like a large positive number in the comparison. So, >>> __pata_platform_probe attempts to use an interrupt when there isn't one. >>> >>> I think the fix would be to change __pata_platform_probe to: >>> >>> if (irq_res && irq_res->start != -1) { >>> >>> but that might have other unintended consequences, so I'll defer to >>> whomever knows more about the intent of this code. >>> >> Something like this patch should work. Thanks for noticing! >> >> drivers/ata/pata_of_platform.c | 2 +- >> 1 files changed, 1 insertions(+), 1 deletions(-) >> >> diff --git a/drivers/ata/pata_of_platform.c b/drivers/ata/pata_of_platform.c >> index 408da30..1f18ad9 100644 >> --- a/drivers/ata/pata_of_platform.c >> +++ b/drivers/ata/pata_of_platform.c >> @@ -52,7 +52,7 @@ static int __devinit pata_of_platform_probe(struct of_device *ofdev, >> >> ret = of_irq_to_resource(dn, 0, &irq_res); >> if (ret == NO_IRQ) >> - irq_res.start = irq_res.end = -1; >> + irq_res.start = irq_res.end = 0; >> else >> irq_res.flags = 0; >> >> > > --------------070500030100010105040908 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Ben Dooks wrote: 
On Mon, Aug 11, 2008 at 07:19:13PM +0400, Anton Vorontsov wrote:
  
When no irq specified, pata_of_platform fills irq_res with -1,
which is wrong to do for two reasons:

1. By definition, 'no irq' should be IRQ 0, not some negative integer;
    

interesting, IRQ 0 is actually valid on some ARM systems.

It is here too, but I believe most of the code uses a virtualized irq number, so physical IRQ 0 would presumably get mapped to a non-zero virtual one.


  

    
2. pata_platform checks for irq_res.start > 0, but since irq_res.start
   is unsigned type, the check will be true for `-1'.

Reported-by: Steven A. Falco <sfalco@harris.com>
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
---

On Mon, Aug 11, 2008 at 10:48:50AM -0400, Steven A. Falco wrote:
    

    

      
I think there is a bug in the communications between pata_of_platform
and pata_platform.  I will refer to the master branch of the DENX git
tree, which is roughly v2.6.26.1 at this time.  I am using a Sequoia
board with a PPC440EPx.

In pata_of_platform, we have:

    ret = of_irq_to_resource(dn, 0, &irq_res);
    if (ret == NO_IRQ)
        irq_res.start = irq_res.end = -1;

so if there is no interrupt defined, then start and end are -1. 
However, __pata_platform_probe has:

    if (irq_res && irq_res->start > 0) {
        irq = irq_res->start;
        irq_flags = irq_res->flags;
    }

You might think that the (irq_res->start > 0) test will fail, as it
should in this no-irq case.  But, start is a u64, so the -1 actually
looks like a large positive number in the comparison.  So,
__pata_platform_probe attempts to use an interrupt when there isn't one.

I think the fix would be to change __pata_platform_probe to:

    if (irq_res && irq_res->start != -1) {

but that might have other unintended consequences, so I'll defer to
whomever knows more about the intent of this code.
      

    

    
Something like this patch should work. Thanks for noticing!

 drivers/ata/pata_of_platform.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/ata/pata_of_platform.c b/drivers/ata/pata_of_platform.c
index 408da30..1f18ad9 100644
--- a/drivers/ata/pata_of_platform.c
+++ b/drivers/ata/pata_of_platform.c
@@ -52,7 +52,7 @@ static int __devinit pata_of_platform_probe(struct of_device *ofdev,
 
 	ret = of_irq_to_resource(dn, 0, &irq_res);
 	if (ret == NO_IRQ)
-		irq_res.start = irq_res.end = -1;
+		irq_res.start = irq_res.end = 0;
 	else
 		irq_res.flags = 0;
 
    

  

  

  








--------------070500030100010105040908--