[patch] enic: fix an endian bug in enic_probe()

[patch] enic: fix an endian bug in enic_probe()

[Dan Carpenter]

"num_vfs" is a u32 but we only use the high 16 bits and the low 16bits
are left as zero.  That isn't a problem for little endian systems but it
will break on big endian ones.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

diff --git a/drivers/net/ethernet/cisco/enic/enic.h b/drivers/net/ethernet/cisco/enic/enic.h
index 99998c6..cf1fb4b 100644
--- a/drivers/net/ethernet/cisco/enic/enic.h
+++ b/drivers/net/ethernet/cisco/enic/enic.h
@@ -94,7 +94,7 @@ struct enic {
 	u32 rx_coalesce_usecs;
 	u32 tx_coalesce_usecs;
 #ifdef CONFIG_PCI_IOV
-	u32 num_vfs;
+	u16 num_vfs;
 #endif
 	struct enic_port_profile *pp;
 
diff --git a/drivers/net/ethernet/cisco/enic/enic_main.c b/drivers/net/ethernet/cisco/enic/enic_main.c
index e27ec1d..9080ed6 100644
--- a/drivers/net/ethernet/cisco/enic/enic_main.c
+++ b/drivers/net/ethernet/cisco/enic/enic_main.c
@@ -2390,7 +2390,7 @@ static int __devinit enic_probe(struct pci_dev *pdev,
 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
 	if (pos) {
 		pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
-			(u16 *)&enic->num_vfs);
+			&enic->num_vfs);
 		if (enic->num_vfs) {
 			err = pci_enable_sriov(pdev, enic->num_vfs);
 			if (err) {

[patch] enic: fix an endian bug in enic_probe()

[Dan Carpenter]

"num_vfs" is a u32 but we only use the high 16 bits and the low 16bits
are left as zero.  That isn't a problem for little endian systems but it
will break on big endian ones.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

diff --git a/drivers/net/ethernet/cisco/enic/enic.h b/drivers/net/ethernet/cisco/enic/enic.h
index 99998c6..cf1fb4b 100644
--- a/drivers/net/ethernet/cisco/enic/enic.h
+++ b/drivers/net/ethernet/cisco/enic/enic.h
@@ -94,7 +94,7 @@ struct enic {
 	u32 rx_coalesce_usecs;
 	u32 tx_coalesce_usecs;
 #ifdef CONFIG_PCI_IOV
-	u32 num_vfs;
+	u16 num_vfs;
 #endif
 	struct enic_port_profile *pp;
 
diff --git a/drivers/net/ethernet/cisco/enic/enic_main.c b/drivers/net/ethernet/cisco/enic/enic_main.c
index e27ec1d..9080ed6 100644
--- a/drivers/net/ethernet/cisco/enic/enic_main.c
+++ b/drivers/net/ethernet/cisco/enic/enic_main.c
@@ -2390,7 +2390,7 @@ static int __devinit enic_probe(struct pci_dev *pdev,
 	pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
 	if (pos) {
 		pci_read_config_word(pdev, pos + PCI_SRIOV_TOTAL_VF,
-			(u16 *)&enic->num_vfs);
+			&enic->num_vfs);
 		if (enic->num_vfs) {
 			err = pci_enable_sriov(pdev, enic->num_vfs);
 			if (err) {

Re: [patch] enic: fix an endian bug in enic_probe()

[David Miller]

From: Dan Carpenter <dan.carpenter@oracle.com>
Date: Thu, 1 Mar 2012 10:19:54 +0300

> "num_vfs" is a u32 but we only use the high 16 bits and the low 16bits
> are left as zero.  That isn't a problem for little endian systems but it
> will break on big endian ones.
> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

Applied.

Re: [patch] enic: fix an endian bug in enic_probe()

[David Miller]

From: Dan Carpenter <dan.carpenter@oracle.com>
Date: Thu, 1 Mar 2012 10:19:54 +0300

> "num_vfs" is a u32 but we only use the high 16 bits and the low 16bits
> are left as zero.  That isn't a problem for little endian systems but it
> will break on big endian ones.
> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

Applied.

RE: [patch] enic: fix an endian bug in enic_probe()

[Christian Benvenuti (benve)]

Thanks Dan, David.

Just one quick comment...
pci_enable_sriov's 2nd input is declared as type int and we
were using u32 instead (for a non negative 16bit value).
With a quick check I noticed that other pci_enable_sriov callers
do something similar and may need to be taken care too:

driver    |  type used
----------+--------------
mlx4      | int
chelsio   | unsigned int
igb       | unsigned int   
igbxe     | unsigned int
emulex    | u32
vxge      | u32


Another option would have been to make all drivers use int
to match pci_enable_sriov (pci_enable_sriov->sriov_enable checks
against negative values).

(BTW, why is pci_enable_sriov prototype using int?)

Thanks
/Chris

> -----Original Message-----
> From: David Miller [mailto:davem@davemloft.net]
> Sent: Thursday, March 01, 2012 2:24 PM
> To: dan.carpenter@oracle.com
> Cc: Roopa Prabhu (roprabhu); Christian Benvenuti (benve); Neel Patel
> (neepatel); Nishank Trivedi (nistrive); netdev@vger.kernel.org;
kernel-
> janitors@vger.kernel.org
> Subject: Re: [patch] enic: fix an endian bug in enic_probe()
> 
> From: Dan Carpenter <dan.carpenter@oracle.com>
> Date: Thu, 1 Mar 2012 10:19:54 +0300
> 
> > "num_vfs" is a u32 but we only use the high 16 bits and the low
> 16bits
> > are left as zero.  That isn't a problem for little endian systems
but
> it
> > will break on big endian ones.
> >
> > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> 
> Applied.

RE: [patch] enic: fix an endian bug in enic_probe()

[Christian Benvenuti (benve)]

Thanks Dan, David.

Just one quick comment...
pci_enable_sriov's 2nd input is declared as type int and we
were using u32 instead (for a non negative 16bit value).
With a quick check I noticed that other pci_enable_sriov callers
do something similar and may need to be taken care too:

driver    |  type used
----------+--------------
mlx4      | int
chelsio   | unsigned int
igb       | unsigned int   
igbxe     | unsigned int
emulex    | u32
vxge      | u32


Another option would have been to make all drivers use int
to match pci_enable_sriov (pci_enable_sriov->sriov_enable checks
against negative values).

(BTW, why is pci_enable_sriov prototype using int?)

Thanks
/Chris

> -----Original Message-----
> From: David Miller [mailto:davem@davemloft.net]
> Sent: Thursday, March 01, 2012 2:24 PM
> To: dan.carpenter@oracle.com
> Cc: Roopa Prabhu (roprabhu); Christian Benvenuti (benve); Neel Patel
> (neepatel); Nishank Trivedi (nistrive); netdev@vger.kernel.org;
kernel-
> janitors@vger.kernel.org
> Subject: Re: [patch] enic: fix an endian bug in enic_probe()
> 
> From: Dan Carpenter <dan.carpenter@oracle.com>
> Date: Thu, 1 Mar 2012 10:19:54 +0300
> 
> > "num_vfs" is a u32 but we only use the high 16 bits and the low
> 16bits
> > are left as zero.  That isn't a problem for little endian systems
but
> it
> > will break on big endian ones.
> >
> > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> 
> Applied.

Re: [patch] enic: fix an endian bug in enic_probe()

[Dan Carpenter]

[-- Attachment #1: Type: text/plain, Size: 936 bytes --]

On Thu, Mar 01, 2012 at 07:23:45PM -0600, Christian Benvenuti (benve) wrote:
> Thanks Dan, David.
> 
> Just one quick comment...
> pci_enable_sriov's 2nd input is declared as type int and we
> were using u32 instead (for a non negative 16bit value).
> With a quick check I noticed that other pci_enable_sriov callers
> do something similar and may need to be taken care too:
> 

The problem was calling pci_read_config_word() not
pci_enable_sriov().  I did consider passing a temporary variable to
pci_read_config_word() and then storing that in num_vfs but it
seemed more complicated.

Looking at the other drivers now, I see that's how some of them do
it.  But it makes sense for them because they have the num_vfs which
the user can configure and the number from the hardware which is the
max that are supported.

Either way would fix it, but I feel that my fix is not terribly
ugly.

regards,
dan carpenter

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [patch] enic: fix an endian bug in enic_probe()

[Dan Carpenter]

[-- Attachment #1: Type: text/plain, Size: 936 bytes --]

On Thu, Mar 01, 2012 at 07:23:45PM -0600, Christian Benvenuti (benve) wrote:
> Thanks Dan, David.
> 
> Just one quick comment...
> pci_enable_sriov's 2nd input is declared as type int and we
> were using u32 instead (for a non negative 16bit value).
> With a quick check I noticed that other pci_enable_sriov callers
> do something similar and may need to be taken care too:
> 

The problem was calling pci_read_config_word() not
pci_enable_sriov().  I did consider passing a temporary variable to
pci_read_config_word() and then storing that in num_vfs but it
seemed more complicated.

Looking at the other drivers now, I see that's how some of them do
it.  But it makes sense for them because they have the num_vfs which
the user can configure and the number from the hardware which is the
max that are supported.

Either way would fix it, but I feel that my fix is not terribly
ugly.

regards,
dan carpenter

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]