[PATCH] scsi_dh_rdac: Changing the memory allocation flag from GFP_KERNEL to GFP_NOIO

[PATCH] scsi_dh_rdac: Changing the memory allocation flag from GFP_KERNEL to GFP_NOIO

[Moger, Babu]

This patch changes the memory allocation flag from GFP_KERNEL to GFP_NOIO which is more appropriate.

This was noticed by Mike Christie and Rob Evers.. 
Mike Christie wrote:
> Rob Evers wrote:
>> +
>> +static int queue_mode_select(struct scsi_device *sdev,
>> +                activate_complete fn, void *data)
>> +{
>> +    struct rdac_queue_data *qdata;
>> +    struct rdac_controller *ctlr;
>> +
>> +    qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
>> +    if (!qdata)
>> +        return SCSI_DH_RETRY;
>> +
>
> It looks like this is called in the main IO path. A failed IO could cause, the path to go down,
> then we would call this to active a new path. If the system needed to write out pages to the same
> disk that just had the failed path so it can allocate memory then GFP_KERNEL would be bad to use here.
> You would use a mempool and still use something like GFP_NOIO. If a mempool is not possible for some 
> reason, then you would want to still use GFP_NOIO.

Using mempool would become bit more complex here because I cannot assume about the number of elements
during the allocation. I would use GFP_NOIO.

CC: Mike Christie <michaelc@cs.wisc.edu>
CC: Rob Evers <revers@redhat.com>
CC: Chandra Seetharaman <chandra.seetharaman@us.ibm.com>
Signed-off-by: Babu Moger <babu.moger@lsi.com>

---
--- linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c.orig	2010-01-05 11:49:22.000000000 -0400
+++ linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c	2010-01-05 11:53:13.000000000 -0400
@@ -624,7 +624,7 @@ static int queue_mode_select(struct scsi
 	struct rdac_queue_data *qdata;
 	struct rdac_controller *ctlr;
 
-	qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
+	qdata = kzalloc(sizeof(*qdata), GFP_NOIO);
 	if (!qdata)
 		return SCSI_DH_RETRY;
 

Re: [dm-devel] [PATCH] scsi_dh_rdac: Changing the memory allocation flag from GFP_KERNEL to GFP_NOIO

[Chandra Seetharaman]

Acked-by: Chandra Seetharaman <sekharan@us.ibm.com>

On Tue, 2010-01-05 at 10:20 -0700, Moger, Babu wrote:
> This patch changes the memory allocation flag from GFP_KERNEL to GFP_NOIO which is more appropriate.
> 
> This was noticed by Mike Christie and Rob Evers.. 
> Mike Christie wrote:
> > Rob Evers wrote:
> >> +
> >> +static int queue_mode_select(struct scsi_device *sdev,
> >> +                activate_complete fn, void *data)
> >> +{
> >> +    struct rdac_queue_data *qdata;
> >> +    struct rdac_controller *ctlr;
> >> +
> >> +    qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
> >> +    if (!qdata)
> >> +        return SCSI_DH_RETRY;
> >> +
> >
> > It looks like this is called in the main IO path. A failed IO could cause, the path to go down,
> > then we would call this to active a new path. If the system needed to write out pages to the same
> > disk that just had the failed path so it can allocate memory then GFP_KERNEL would be bad to use here.
> > You would use a mempool and still use something like GFP_NOIO. If a mempool is not possible for some 
> > reason, then you would want to still use GFP_NOIO.
> 
> Using mempool would become bit more complex here because I cannot assume about the number of elements
> during the allocation. I would use GFP_NOIO.
> 
> CC: Mike Christie <michaelc@cs.wisc.edu>
> CC: Rob Evers <revers@redhat.com>
> CC: Chandra Seetharaman <chandra.seetharaman@us.ibm.com>
> Signed-off-by: Babu Moger <babu.moger@lsi.com>
> 
> ---
> --- linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c.orig	2010-01-05 11:49:22.000000000 -0400
> +++ linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c	2010-01-05 11:53:13.000000000 -0400
> @@ -624,7 +624,7 @@ static int queue_mode_select(struct scsi
>  	struct rdac_queue_data *qdata;
>  	struct rdac_controller *ctlr;
> 
> -	qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
> +	qdata = kzalloc(sizeof(*qdata), GFP_NOIO);
>  	if (!qdata)
>  		return SCSI_DH_RETRY;
> 
> 
> 
> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel

Re: scsi_dh_rdac: Changing the memory allocation flag from GFP_KERNEL to GFP_NOIO

[Mike Snitzer]

This patch appears to have gotten lost in the shuffle but still seems
relevant.  I found it while auditing dm-devel's patchwork, patch is
here: https://patchwork.kernel.org/patch/71073/

On Tue, Jan 05 2010 at 12:20pm -0500,
Moger, Babu <Babu.Moger@lsi.com> wrote:

> This patch changes the memory allocation flag from GFP_KERNEL to GFP_NOIO which is more appropriate.
> 
> This was noticed by Mike Christie and Rob Evers.. 
> Mike Christie wrote:
> > Rob Evers wrote:
> >> +
> >> +static int queue_mode_select(struct scsi_device *sdev,
> >> +                activate_complete fn, void *data)
> >> +{
> >> +    struct rdac_queue_data *qdata;
> >> +    struct rdac_controller *ctlr;
> >> +
> >> +    qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
> >> +    if (!qdata)
> >> +        return SCSI_DH_RETRY;
> >> +
> >
> > It looks like this is called in the main IO path. A failed IO could cause, the path to go down,
> > then we would call this to active a new path. If the system needed to write out pages to the same
> > disk that just had the failed path so it can allocate memory then GFP_KERNEL would be bad to use here.
> > You would use a mempool and still use something like GFP_NOIO. If a mempool is not possible for some 
> > reason, then you would want to still use GFP_NOIO.
> 
> Using mempool would become bit more complex here because I cannot assume about the number of elements
> during the allocation. I would use GFP_NOIO.
> 
> CC: Mike Christie <michaelc@cs.wisc.edu>
> CC: Rob Evers <revers@redhat.com>
> CC: Chandra Seetharaman <chandra.seetharaman@us.ibm.com>
> Signed-off-by: Babu Moger <babu.moger@lsi.com>
> 
> ---
> --- linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c.orig	2010-01-05 11:49:22.000000000 -0400
> +++ linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c	2010-01-05 11:53:13.000000000 -0400
> @@ -624,7 +624,7 @@ static int queue_mode_select(struct scsi
>  	struct rdac_queue_data *qdata;
>  	struct rdac_controller *ctlr;
>  
> -	qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
> +	qdata = kzalloc(sizeof(*qdata), GFP_NOIO);
>  	if (!qdata)
>  		return SCSI_DH_RETRY;
>  
> 
> 
> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel

Re: scsi_dh_rdac: Changing the memory allocation flag from GFP_KERNEL to GFP_NOIO

[Moger, Babu]

> -----Original Message-----
> From: Mike Snitzer [mailto:snitzer@redhat.com]
> Sent: Thursday, November 04, 2010 10:15 AM
> To: device-mapper development
> Cc: linux-scsi@vger.kernel.org; Chandra Seetharaman;
> michaelc@cs.wisc.edu; Rob Evers; Moger, Babu
> Subject: Re: scsi_dh_rdac: Changing the memory allocation flag from
> GFP_KERNEL to GFP_NOIO
> 
> This patch appears to have gotten lost in the shuffle but still seems
> relevant.  I found it while auditing dm-devel's patchwork, patch is
> here: https://patchwork.kernel.org/patch/71073/

  Yes, Please.. We need this patch.. 
> 
> On Tue, Jan 05 2010 at 12:20pm -0500,
> Moger, Babu <Babu.Moger@lsi.com> wrote:
> 
> > This patch changes the memory allocation flag from GFP_KERNEL to
> GFP_NOIO which is more appropriate.
> >
> > This was noticed by Mike Christie and Rob Evers..
> > Mike Christie wrote:
> > > Rob Evers wrote:
> > >> +
> > >> +static int queue_mode_select(struct scsi_device *sdev,
> > >> +                activate_complete fn, void *data)
> > >> +{
> > >> +    struct rdac_queue_data *qdata;
> > >> +    struct rdac_controller *ctlr;
> > >> +
> > >> +    qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
> > >> +    if (!qdata)
> > >> +        return SCSI_DH_RETRY;
> > >> +
> > >
> > > It looks like this is called in the main IO path. A failed IO could
> cause, the path to go down,
> > > then we would call this to active a new path. If the system needed
> to write out pages to the same
> > > disk that just had the failed path so it can allocate memory then
> GFP_KERNEL would be bad to use here.
> > > You would use a mempool and still use something like GFP_NOIO. If a
> mempool is not possible for some
> > > reason, then you would want to still use GFP_NOIO.
> >
> > Using mempool would become bit more complex here because I cannot
> assume about the number of elements
> > during the allocation. I would use GFP_NOIO.
> >
> > CC: Mike Christie <michaelc@cs.wisc.edu>
> > CC: Rob Evers <revers@redhat.com>
> > CC: Chandra Seetharaman <chandra.seetharaman@us.ibm.com>
> > Signed-off-by: Babu Moger <babu.moger@lsi.com>
> >
> > ---
> > --- linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c.orig
> 	2010-01-05 11:49:22.000000000 -0400
> > +++ linux-2.6.33-rc2/drivers/scsi/device_handler/scsi_dh_rdac.c	2010-
> 01-05 11:53:13.000000000 -0400
> > @@ -624,7 +624,7 @@ static int queue_mode_select(struct scsi
> >  	struct rdac_queue_data *qdata;
> >  	struct rdac_controller *ctlr;
> >
> > -	qdata = kzalloc(sizeof(*qdata), GFP_KERNEL);
> > +	qdata = kzalloc(sizeof(*qdata), GFP_NOIO);
> >  	if (!qdata)
> >  		return SCSI_DH_RETRY;
> >
> >
> >
> >
> > --
> > dm-devel mailing list
> > dm-devel@redhat.com
> > https://www.redhat.com/mailman/listinfo/dm-devel