[ PATCH ] mptsas: Fix oops for insmod during kexec

[ PATCH ] mptsas: Fix oops for insmod during kexec

[Judith Lebzelter]

Hello,

This patch is to fix an oops on insmod for mptsas during kexec.
This applies to 2.6.21-rc3.

Signed-off-by:  Judith Lebzelter <judith@osdl.org

---


Index: linux-2.6.21-rc3/drivers/message/fusion/mptsas.c
===================================================================
--- linux-2.6.21-rc3.orig/drivers/message/fusion/mptsas.c
+++ linux-2.6.21-rc3/drivers/message/fusion/mptsas.c
@@ -815,7 +815,7 @@ mptsas_taskmgmt_complete(MPT_ADAPTER *io
 static int
 mptsas_ioc_reset(MPT_ADAPTER *ioc, int reset_phase)
 {
-	MPT_SCSI_HOST	*hd = (MPT_SCSI_HOST *)ioc->sh->hostdata;
+	MPT_SCSI_HOST	*hd;
 	struct mptsas_target_reset_event *target_reset_list, *n;
 	int rc;
 
@@ -827,7 +827,10 @@ mptsas_ioc_reset(MPT_ADAPTER *ioc, int r
 	if (reset_phase != MPT_IOC_POST_RESET)
 		goto out;
 
-	if (!hd || !hd->ioc)
+	if (!ioc->sh || !ioc->sh->hostdata)
+		goto out;
+	hd = (MPT_SCSI_HOST *)ioc->sh->hostdata;
+	if (!hd->ioc)
 		goto out;
 
 	if (list_empty(&hd->target_reset_list))

Re: [ PATCH ] mptsas: Fix oops for insmod during kexec

[Christoph Hellwig]

On Fri, Mar 09, 2007 at 01:07:44PM -0800, Judith Lebzelter wrote:
> Hello,
> 
> This patch is to fix an oops on insmod for mptsas during kexec.
> This applies to 2.6.21-rc3.
> 
> Signed-off-by:  Judith Lebzelter <judith@osdl.org
> 
> ---
> 
> 
> Index: linux-2.6.21-rc3/drivers/message/fusion/mptsas.c
> ===================================================================
> --- linux-2.6.21-rc3.orig/drivers/message/fusion/mptsas.c
> +++ linux-2.6.21-rc3/drivers/message/fusion/mptsas.c
> @@ -815,7 +815,7 @@ mptsas_taskmgmt_complete(MPT_ADAPTER *io
>  static int
>  mptsas_ioc_reset(MPT_ADAPTER *ioc, int reset_phase)
>  {
> -	MPT_SCSI_HOST	*hd = (MPT_SCSI_HOST *)ioc->sh->hostdata;
> +	MPT_SCSI_HOST	*hd;
>  	struct mptsas_target_reset_event *target_reset_list, *n;
>  	int rc;
>  
> @@ -827,7 +827,10 @@ mptsas_ioc_reset(MPT_ADAPTER *ioc, int r
>  	if (reset_phase != MPT_IOC_POST_RESET)
>  		goto out;
>  
> -	if (!hd || !hd->ioc)
> +	if (!ioc->sh || !ioc->sh->hostdata)
> +		goto out;
> +	hd = (MPT_SCSI_HOST *)ioc->sh->hostdata;
> +	if (!hd->ioc)

This needs a much better explanation.  Please show the codepath
leading to this.

RE: [ PATCH ] mptsas: Fix oops for insmod during kexec

[Moore, Eric]

On Friday, March 09, 2007 2:23 PM, Christoph Hellwig wrote: 


> This needs a much better explanation.  Please show the codepath
> leading to this.
> 
>

I was working with Judith on this patch earlier this week, so this is
approved by LSI,,, ACK.

This fix's an oops during driver load time.   mptsas_probe calls
mpt_attach(over in mptbase.c).  Inside that call, we read some
manufacturing config pages to setup some defaults.  While reading the
config pages, the firmware doesn't complete the reply in time, and we
have a timeout. The timeout results in hardreset handler being called.
The hardreset handler calls all the fusion upper layer driver reset
callback handlers.  The mptsas_ioc_reset function is the callback
handler in mptsas.c.   So where I'm getting to, is mptsas_ioc_reset is
getting called before scsi_host_alloc is called, and the pointer ioc->sh
is NULL.,,, as well as the hostdata. 
 

Re: [ PATCH ] mptsas: Fix oops for insmod during kexec

[Christoph Hellwig]

On Fri, Mar 09, 2007 at 03:22:13PM -0700, Moore, Eric wrote:
> On Friday, March 09, 2007 2:23 PM, Christoph Hellwig wrote: 
> 
> 
> > This needs a much better explanation.  Please show the codepath
> > leading to this.
> > 
> >
> 
> I was working with Judith on this patch earlier this week, so this is
> approved by LSI,,, ACK.
> 
> This fix's an oops during driver load time.   mptsas_probe calls
> mpt_attach(over in mptbase.c).  Inside that call, we read some
> manufacturing config pages to setup some defaults.  While reading the
> config pages, the firmware doesn't complete the reply in time, and we
> have a timeout. The timeout results in hardreset handler being called.
> The hardreset handler calls all the fusion upper layer driver reset
> callback handlers.  The mptsas_ioc_reset function is the callback
> handler in mptsas.c.   So where I'm getting to, is mptsas_ioc_reset is
> getting called before scsi_host_alloc is called, and the pointer ioc->sh
> is NULL.,,, as well as the hostdata. 

Ok.  Please put a reformulated version of the above statement as a comment
above the checks, so that people don't rip it out again.

[ PATCH ] mptsas: Fix oops during driver load time(rev 2)

[Judith Lebzelter]

This fixes an oops during driver load time.   

mptsas_probe calls mpt_attach(over in mptbase.c).  Inside that 
call, we read some manufacturing config pages to setup some 
defaults.  While reading the config pages, the firmware doesn't 
complete the reply in time, and we have a timeout. The timeout 
results in hardreset handler being called.  The hardreset 
handler calls all the fusion upper layer driver reset callback 
handlers.  The mptsas_ioc_reset function is the callback handler 
in mptsas.c.   In summary, mptsas_ioc_reset is getting called 
before scsi_host_alloc is called, and the pointer ioc->sh is 
NULL, as well as the hostdata.

Signed-off-by:  Judith Lebzelter <judith@osdl.org

---
Sorry I was not more descriptive.  Here is the patch with Eric's 
description as requested.

Index: linux-2.6.21-rc3/drivers/message/fusion/mptsas.c
===================================================================
--- linux-2.6.21-rc3.orig/drivers/message/fusion/mptsas.c
+++ linux-2.6.21-rc3/drivers/message/fusion/mptsas.c
@@ -815,7 +815,7 @@ mptsas_taskmgmt_complete(MPT_ADAPTER *io
 static int
 mptsas_ioc_reset(MPT_ADAPTER *ioc, int reset_phase)
 {
-	MPT_SCSI_HOST	*hd = (MPT_SCSI_HOST *)ioc->sh->hostdata;
+	MPT_SCSI_HOST	*hd;
 	struct mptsas_target_reset_event *target_reset_list, *n;
 	int rc;
 
@@ -827,7 +827,10 @@ mptsas_ioc_reset(MPT_ADAPTER *ioc, int r
 	if (reset_phase != MPT_IOC_POST_RESET)
 		goto out;
 
-	if (!hd || !hd->ioc)
+	if (!ioc->sh || !ioc->sh->hostdata)
+		goto out;
+	hd = (MPT_SCSI_HOST *)ioc->sh->hostdata;
+	if (!hd->ioc)
 		goto out;
 
 	if (list_empty(&hd->target_reset_list))