Skip to contentSkip to conversation

hdlc: fix compile problem

PATCH In review Stephen Hemminger 3 participants 8 messages Jan 7, 2009 – Jan 8, 2009

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hdlc: fix compile problem

Fix build problem with hdlc driver. hdlc_null_ops was being referenced before it was defined!

drivers/net/wan/hdlc.c +2 -2
@@ -40,6 +40,8 @@
 
 static const char* version = "HDLC support module revision 1.22";
 
+static const struct header_ops hdlc_null_ops;
+
 #undef DEBUG_LINK
 
 static struct hdlc_proto *first_proto;
@@ -218,8 +220,6 @@ int hdlc_ioctl(struct net_device *dev, s
 	return -EINVAL;
 }
 
-static const struct header_ops hdlc_null_ops;
-
 static void hdlc_setup_dev(struct net_device *dev)
 {
 	/* Re-init all variables changed by HDLC protocol drivers,
maintainer
#1 lore raw
Quoted 2 lines · Stephen Hemminger

Fix build problem with hdlc driver. hdlc_null_ops was being referenced before it was defined!

:-)

Unfortunately I can't see it, which tree is it?

In David's and Linus' trees it appears twice, it seem to be entirely correct and btw you are the author :-)
... and it compiled here 100 times or so :-)

static const struct header_ops hdlc_null_ops;
                               ^^^^^^^^^^^^^
static void hdlc_setup_dev(struct net_device *dev)
{
        /* Re-init all variables changed by HDLC protocol drivers,
         * including ether_setup() called from hdlc_raw_eth.c.
         */
        dev->get_stats           = hdlc_get_stats;
        dev->flags               = IFF_POINTOPOINT | IFF_NOARP;
        dev->mtu                 = HDLC_MAX_MTU;
        dev->type                = ARPHRD_RAWHDLC;
        dev->hard_header_len     = 16;
        dev->addr_len            = 0;
        dev->header_ops          = &hdlc_null_ops;
                                  ^^^^^^^^^^^^^^^^
        dev->change_mtu          = hdlc_change_mtu;
}
Signature · 1 line
Krzysztof Halasa
#2 lore raw
Quoted 23 lines · On Thu, 08 Jan 2009 00:21:03 +0100 Krzysztof Halasa <[email protected]> wrote:
Quoted 2 lines · Stephen Hemminger

Fix build problem with hdlc driver. hdlc_null_ops was being referenced before it was defined!

:-)

Unfortunately I can't see it, which tree is it?

In David's and Linus' trees it appears twice, it seem to be entirely correct and btw you are the author :-)
... and it compiled here 100 times or so :-)

static const struct header_ops hdlc_null_ops;
                               ^^^^^^^^^^^^^
static void hdlc_setup_dev(struct net_device *dev)
{
        /* Re-init all variables changed by HDLC protocol drivers,
         * including ether_setup() called from hdlc_raw_eth.c.
         */
        dev->get_stats           = hdlc_get_stats;
        dev->flags               = IFF_POINTOPOINT | IFF_NOARP;
Quoted 8 lines
        dev->mtu                 = HDLC_MAX_MTU;
        dev->type                = ARPHRD_RAWHDLC;
        dev->hard_header_len     = 16;
        dev->addr_len            = 0;
        dev->header_ops          = &hdlc_null_ops;
                                  ^^^^^^^^^^^^^^^^
        dev->change_mtu          = hdlc_change_mtu;
}

It was introduced (by me) in previous patch. Since that isn't applied yet, I'll go back and fix.

maintainer
#3 lore raw
Quoted 2 lines · Stephen Hemminger

It was introduced (by me) in previous patch. Since that isn't applied yet, I'll go back and fix.

I can see it now, I must have missed it because of the "synclink" subject.

@@ -106,7 +98,7 @@ static int hdlc_device_event(struct noti

        if (dev_net(dev) != &init_net)
                return NOTIFY_DONE;

- if (dev->get_stats != hdlc_get_stats)
+ if (dev->header_ops != &hdlc_null_ops)
return NOTIFY_DONE; /* not an HDLC device */

This won't work because hdlc_null_ops may be substituted by cisco_header_ops or ppp_header_ops. Current test works for now but is wrong, too, at least for wanxl driver (it tries to use its get_stats()).

Perhaps some
+ if (dev->header_ops->something != &something)
would do? Alternatively we can return to hdlc_carrier_on|off() and hdlc_device_event() won't be needed.

Essentially, it's a bit hard to check if the device is one of ours if we are the middle layer.

There are two layers here:
- hardware drivers (10+)
- protocol drivers (5 - hdlc* files)

Protocol drivers want to define things like:
- dev->header_ops (no problem)
- dev->change_mtu (perhaps)
- dev->hard_start_xmit

Hardware drivers need to define the rest of NDO.

Hmm, not much, only hard_start_xmit() is really problematic. I think we need a trampoline: hw driver will define the NDO and set hard_start_xmit and possibly change_mtu to (exported by hdlc.c) hdlc_xmit (hdlc_change_mtu), and hdlc_start_xmit will just call protocol's xmit().
We could also use it for the test in hdlc_device_event:

	if (dev->netdev_ops->start_hard_xmit != hdlc_xmit)

If we could then get rid of this special xmit() and do all header and TX preparations in hard_header(), it would be even better (cleaner) though IIRC it requires some changes to Ethernet(?) code.

The changes have to be applied to the 4 synclink drivers and to most .c files in drivers/net/wan simultaneously, perhaps it's more practical if I do it?

Signature · 1 line
Krzysztof Halasa
#4 lore raw
Quoted 49 lines · On Thu, 08 Jan 2009 02:00:00 +0100 Krzysztof Halasa <[email protected]> wrote:
Quoted 2 lines · Stephen Hemminger

It was introduced (by me) in previous patch. Since that isn't applied yet, I'll go back and fix.

I can see it now, I must have missed it because of the "synclink" subject.

(quoted diff) +1 -1
@@ -106,7 +98,7 @@ static int hdlc_device_event(struct noti
        if (dev_net(dev) != &init_net)
                return NOTIFY_DONE;
 
-       if (dev->get_stats != hdlc_get_stats)
+       if (dev->header_ops != &hdlc_null_ops)
                return NOTIFY_DONE; /* not an HDLC device */
 

This won't work because hdlc_null_ops may be substituted by cisco_header_ops or ppp_header_ops. Current test works for now but is wrong, too, at least for wanxl driver (it tries to use its get_stats()).

Perhaps some
+ if (dev->header_ops->something != &something)
would do? Alternatively we can return to hdlc_carrier_on|off() and hdlc_device_event() won't be needed.

Essentially, it's a bit hard to check if the device is one of ours if we are the middle layer.

There are two layers here:
- hardware drivers (10+)
- protocol drivers (5 - hdlc* files)

Protocol drivers want to define things like:
- dev->header_ops (no problem)
- dev->change_mtu (perhaps)
- dev->hard_start_xmit

Hardware drivers need to define the rest of NDO.

Hmm, not much, only hard_start_xmit() is really problematic. I think we need a trampoline: hw driver will define the NDO and set hard_start_xmit and possibly change_mtu to (exported by hdlc.c) hdlc_xmit (hdlc_change_mtu), and hdlc_start_xmit will just call protocol's xmit().
We could also use it for the test in hdlc_device_event:

	if (dev->netdev_ops->start_hard_xmit != hdlc_xmit)
Quoted 8 lines

If we could then get rid of this special xmit() and do all header and TX preparations in hard_header(), it would be even better (cleaner) though IIRC it requires some changes to Ethernet(?) code.

The changes have to be applied to the 4 synclink drivers and to most .c files in drivers/net/wan simultaneously, perhaps it's more practical if I do it?

Alternatively, how about this:

drivers/net/wan/hdlc.c +4 -3
@@ -40,8 +40,6 @@
 
 static const char* version = "HDLC support module revision 1.22";
 
-static const struct header_ops hdlc_null_ops;
-
 #undef DEBUG_LINK
 
 static struct hdlc_proto *first_proto;
@@ -100,7 +98,7 @@ static int hdlc_device_event(struct noti
 	if (dev_net(dev) != &init_net)
 		return NOTIFY_DONE;
 
-	if (dev->header_ops != &hdlc_null_ops)
+	if (dev->priv_flags & IFF_WAN_HDLC)
 		return NOTIFY_DONE; /* not an HDLC device */
 
 	if (event != NETDEV_CHANGE)
@@ -220,12 +218,15 @@ int hdlc_ioctl(struct net_device *dev, s
 	return -EINVAL;
 }
 
+static const struct header_ops hdlc_null_ops;
+
 static void hdlc_setup_dev(struct net_device *dev)
 {
 	/* Re-init all variables changed by HDLC protocol drivers,
 	 * including ether_setup() called from hdlc_raw_eth.c.
 	 */
 	dev->flags		 = IFF_POINTOPOINT | IFF_NOARP;
+	dev->priv_flags		 = IFF_WAN_HDLC;
 	dev->mtu		 = HDLC_MAX_MTU;
 	dev->type		 = ARPHRD_RAWHDLC;
 	dev->hard_header_len	 = 16;
include/linux/if.h +1 -0
@@ -66,6 +66,7 @@
 #define IFF_SLAVE_NEEDARP 0x40		/* need ARPs for validation	*/
 #define IFF_ISATAP	0x80		/* ISATAP interface (RFC4214)	*/
 #define IFF_MASTER_ARPMON 0x100		/* bonding master, ARP mon in use */
+#define IFF_WAN_HDLC	0x200		/* WAN HDLC device              */
 
 #define IF_GET_IFACE	0x0001		/* for querying only */
 #define IF_GET_PROTO	0x0002
maintainer
#5 lore raw
Quoted 14 lines · Stephen Hemminger

Alternatively, how about this:

drivers/net/wan/hdlc.c +0 -2
@@ -40,8 +40,6 @@
 
 static const char* version = "HDLC support module revision 1.22";
 
-static const struct header_ops hdlc_null_ops;
-
 #undef DEBUG_LINK
 
 static struct hdlc_proto *first_proto;

drivers/net/wan/hdlc.c static int hdlc_device_event(struct noti · line 104

Quoted 8 lines
drivers/net/wan/hdlc.c +1 -1
@@ -100,7 +98,7 @@ static int hdlc_device_event(struct noti
 	if (dev_net(dev) != &init_net)
 		return NOTIFY_DONE;
 
-	if (dev->header_ops != &hdlc_null_ops)
+	if (dev->priv_flags & IFF_WAN_HDLC)
 		return NOTIFY_DONE; /* not an HDLC device */
 
 	if (event != NETDEV_CHANGE)

"dev->netdev_ops->start_hard_xmit != hdlc_xmit" is a dirty hack. IFF_WAN_HDLC is much cleaner though it's an additional bit consumed. Sure, I like it.

Signature · 1 line
Krzysztof Halasa
maintainer
#6 lore raw

From: Krzysztof Halasa <[email protected]>
Date: Thu, 08 Jan 2009 02:26:31 +0100

Quoted 3 lines

"dev->netdev_ops->start_hard_xmit != hdlc_xmit" is a dirty hack. IFF_WAN_HDLC is much cleaner though it's an additional bit consumed. Sure, I like it.

Ok, Stephen please formally resubmit this final version with full commit message etc..

Thanks.

maintainer
#7 lore raw
In reply to Stephen Hemminger
Quoted 3 lines
Quoted 3 lines

The changes have to be applied to the 4 synclink drivers and to most .c files in drivers/net/wan simultaneously, perhaps it's more practical if I do it?

Ok, I will do.

Signature · 1 line
Krzysztof Halasa