Gigabit Kconfig problems with yesterday's update

Gigabit Kconfig problems with yesterday's update

[walt]

I have one machine with a gigabit NIC which I updated today from Linus'
bk tree.

The problem is that I was not asked if I wanted the 'new' gigabit
support and therefore the tg3 support was dropped from my new .config.

I edited .config by hand and deleted any mention of ethernet support --
and only then did 'make oldconfig' ask me the right questions.

Also: the phrase (10 or 100Mbit) should be deleted from the 'Ethernet'
menu item since it implies (wrongly) that the item is not needed for
gigabit support.


Thanks!

Re: Gigabit Kconfig problems with yesterday's update

[Adrian Bunk]

On Sat, May 29, 2004 at 07:51:41AM -0700, walt wrote:
> I have one machine with a gigabit NIC which I updated today from Linus'
> bk tree.
> 
> The problem is that I was not asked if I wanted the 'new' gigabit
> support and therefore the tg3 support was dropped from my new .config.
> 
> I edited .config by hand and deleted any mention of ethernet support --
> and only then did 'make oldconfig' ask me the right questions.
>...

Please send your old .config .

> Thanks!

TIA
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

[patch] Re: Gigabit Kconfig problems with yesterday's update

[Adrian Bunk]

On Sat, May 29, 2004 at 07:51:41AM -0700, walt wrote:

> I have one machine with a gigabit NIC which I updated today from Linus'
> bk tree.
> 
> The problem is that I was not asked if I wanted the 'new' gigabit
> support and therefore the tg3 support was dropped from my new .config.
> 
> I edited .config by hand and deleted any mention of ethernet support --
> and only then did 'make oldconfig' ask me the right questions.
> 
> Also: the phrase (10 or 100Mbit) should be deleted from the 'Ethernet'
> menu item since it implies (wrongly) that the item is not needed for
> gigabit support.

Thanks for this good report (and I'm now understanding the issue without 
requiring your .config).

@Jeff:
At a first glance, it seems the patch below that simply removes the 
dependency of NET_GIGE on NET_ETHERNET would suffice.

Is this correct or did I miss something?

> Thanks!

cu
Adrian

--- linux-2.6.7-rc2-full/drivers/net/Kconfig.old	2004-05-30 15:33:24.000000000 +0200
+++ linux-2.6.7-rc2-full/drivers/net/Kconfig	2004-05-30 15:38:41.000000000 +0200
@@ -1879,7 +1879,7 @@
 
 config NET_GIGE
 	bool "Gigabit Ethernet (1000/10000 Mbit) controller support"
-	depends on NETDEVICES && NET_ETHERNET && (PCI || SBUS)
+	depends on NETDEVICES && (PCI || SBUS)
 	help
 	  Gigabit ethernet.  It's yummy and fast, fast, fast.
 

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Danny ter Haar]

Quoting Adrian Bunk (bunk@fs.tum.de):
> @Jeff:
> At a first glance, it seems the patch below that simply removes the 
> dependency of NET_GIGE on NET_ETHERNET would suffice.
> 
> Is this correct or did I miss something?
> 
> cu
> Adrian
> 
> --- linux-2.6.7-rc2-full/drivers/net/Kconfig.old	2004-05-30 15:33:24.000000000 +0200
> +++ linux-2.6.7-rc2-full/drivers/net/Kconfig	2004-05-30 15:38:41.000000000 +0200
> @@ -1879,7 +1879,7 @@
>  
>  config NET_GIGE
>  	bool "Gigabit Ethernet (1000/10000 Mbit) controller support"
> -	depends on NETDEVICES && NET_ETHERNET && (PCI || SBUS)
> +	depends on NETDEVICES && (PCI || SBUS)
>  	help
>  	  Gigabit ethernet.  It's yummy and fast, fast, fast.

Fresh source, this patch, old config from 2.6.7-rc1-bk2 and make
oldconfig now works ;)

Zanks

Danny
-- 
"If Microsoft had been the innovative company that it calls itself, it 
would have taken the opportunity to take a radical leap beyond the Mac, 
instead of producing a feeble, me-too implementation." - Douglas Adams -

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Randy.Dunlap]

On Sun, 30 May 2004 16:37:34 +0200 Danny ter Haar <dth@dth.net> wrote:

| Quoting Adrian Bunk (bunk@fs.tum.de):
| > @Jeff:
| > At a first glance, it seems the patch below that simply removes the 
| > dependency of NET_GIGE on NET_ETHERNET would suffice.
| > 
| > Is this correct or did I miss something?
| > 
| > cu
| > Adrian
| > 
| > --- linux-2.6.7-rc2-full/drivers/net/Kconfig.old	2004-05-30 15:33:24.000000000 +0200
| > +++ linux-2.6.7-rc2-full/drivers/net/Kconfig	2004-05-30 15:38:41.000000000 +0200
| > @@ -1879,7 +1879,7 @@
| >  
| >  config NET_GIGE
| >  	bool "Gigabit Ethernet (1000/10000 Mbit) controller support"
| > -	depends on NETDEVICES && NET_ETHERNET && (PCI || SBUS)
| > +	depends on NETDEVICES && (PCI || SBUS)
| >  	help
| >  	  Gigabit ethernet.  It's yummy and fast, fast, fast.
| 
| Fresh source, this patch, old config from 2.6.7-rc1-bk2 and make
| oldconfig now works ;)

I looked at this last night, after the first reported problem

I'm not surprised that this works, but I'm concerned about the use
of CONFIG_NET_ETHERNET in net/ipv4/arp.c.  I'd _guess_ that
CONFIG_NET_ETHERNET needs to be seen as enabled over there,
even though it's not enabled in drivers/net/*.

I don't know if it's a satisfactory solution, but
net/ipv4/arp.c could also check for CONFIG_NET_GIGE.
However, that doesn't seem like the right thing to do IMO.

--
~Randy

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Jeff Garzik]

[-- Attachment #1: Type: text/plain, Size: 133 bytes --]

NET_GIGE is rightly dependent on NET_ETHERNET, as it is a subset.

I wonder if the attached patch "fixes" peoples config problems :)

[-- Attachment #2: patch --]
[-- Type: text/plain, Size: 457 bytes --]

===== drivers/net/Kconfig 1.74 vs edited =====
--- 1.74/drivers/net/Kconfig	2004-05-27 16:42:40 -04:00
+++ edited/drivers/net/Kconfig	2004-05-30 13:49:48 -04:00
@@ -163,7 +163,7 @@
 	depends on NETDEVICES
 
 config NET_ETHERNET
-	bool "Ethernet (10 or 100Mbit)"
+	bool "Ethernet (10/100/1000/10000 Mbit)"
 	---help---
 	  Ethernet (also called IEEE 802.3 or ISO 8802-2) is the most common
 	  type of Local Area Network (LAN) in universities and companies.

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Adrian Bunk]

On Sun, May 30, 2004 at 01:51:33PM -0400, Jeff Garzik wrote:

> NET_GIGE is rightly dependent on NET_ETHERNET, as it is a subset.
> 
> I wonder if the attached patch "fixes" peoples config problems :)
> 
> ===== drivers/net/Kconfig 1.74 vs edited =====
> --- 1.74/drivers/net/Kconfig	2004-05-27 16:42:40 -04:00
> +++ edited/drivers/net/Kconfig	2004-05-30 13:49:48 -04:00
> @@ -163,7 +163,7 @@
>  	depends on NETDEVICES
>  
>  config NET_ETHERNET
> -	bool "Ethernet (10 or 100Mbit)"
> +	bool "Ethernet (10/100/1000/10000 Mbit)"
>  	---help---
>  	  Ethernet (also called IEEE 802.3 or ISO 8802-2) is the most common
>  	  type of Local Area Network (LAN) in universities and companies.


This is not sufficient since it's still under the
"Ethernet (10 or 100Mbit)" menu.

What about the patch below?

cu
Adrian

--- linux-2.6.7-rc2-full/drivers/net/Kconfig.old	2004-05-30 21:29:46.000000000 +0200
+++ linux-2.6.7-rc2-full/drivers/net/Kconfig	2004-05-30 21:34:42.000000000 +0200
@@ -159,11 +159,8 @@
 #	Ethernet
 #
 
-menu "Ethernet (10 or 100Mbit)"
-	depends on NETDEVICES
-
 config NET_ETHERNET
-	bool "Ethernet (10 or 100Mbit)"
+	bool "Ethernet (10/100/1000/10000 Mbit)"
 	---help---
 	  Ethernet (also called IEEE 802.3 or ISO 8802-2) is the most common
 	  type of Local Area Network (LAN) in universities and companies.
@@ -188,6 +185,9 @@
 	  kernel: saying N will just cause the configurator to skip all
 	  the questions about Ethernet network cards. If unsure, say N.
 
+menu "Ethernet (10/100 Mbit)"
+	depends on NET_ETHERNET
+
 config MII
 	tristate "Generic Media Independent Interface device support"
 	depends on NET_ETHERNET
@@ -1875,7 +1875,7 @@
 #
 
 menu "Gigabit Ethernet (1000/10000 Mbit)"
-	depends on NETDEVICES
+	depends on NET_ETHERNET
 
 config NET_GIGE
 	bool "Gigabit Ethernet (1000/10000 Mbit) controller support"

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Linus Torvalds]

On Sun, 30 May 2004, Adrian Bunk wrote:
> 
> 
> This is not sufficient since it's still under the
> "Ethernet (10 or 100Mbit)" menu.
> 
> What about the patch below?
>
>  config NET_ETHERNET
> -	bool "Ethernet (10 or 100Mbit)"
> +	bool "Ethernet (10/100/1000/10000 Mbit)"

Ok, this part is obviously sane.

However:

> +menu "Ethernet (10/100 Mbit)"
> +	depends on NET_ETHERNET
> +
>  
>  menu "Gigabit Ethernet (1000/10000 Mbit)"
> -	depends on NETDEVICES
> +	depends on NET_ETHERNET

is it really sane these days to split out gigabit from the "regular" 
ethernets? gigabit ethernet is getting quite a bit more common than it 
used to be, and a lot of the gigabit devices are just "standard ethernet" 
as far as the user is concerned, and in fact they are often _used_ in just 
regular 10/100Mbps setups.

In other words: it's quite understandable if somebody doesn't even
_realize_ that his chip supports gigabit speeds these days.

So wouldn't the right fix be to just remove the distinction between 
"regular" and "gigabit" ethernet, and put them all in the same menu?

If the menu then grows really big, maybe we can split it according to some 
_saner_ split. Not that I can see off-hand what that would be, but..

		Linus

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Jeff Garzik]

Linus Torvalds wrote:
> is it really sane these days to split out gigabit from the "regular" 
> ethernets? gigabit ethernet is getting quite a bit more common than it 
> used to be, and a lot of the gigabit devices are just "standard ethernet" 
> as far as the user is concerned, and in fact they are often _used_ in just 
> regular 10/100Mbps setups.

> In other words: it's quite understandable if somebody doesn't even
> _realize_ that his chip supports gigabit speeds these days.

That's ok.  Most users have no idea that their "Surecom 1234 Zippy 
Ethernet" is really an 8139 chip, and that they need the 8139too driver.

Kernel configuration is for kernel developers, kernel vendors 
(distributions), and power users, all of whom are expected to know what 
they are doing.

So, I disagree with that argument.

However...

> If the menu then grows really big, maybe we can split it according to some 
> _saner_ split. Not that I can see off-hand what that would be, but..

...I won't reject arguments based on general annoyance, or specific 
annoyance at the categorizing.  This change was the way I preferred that 
the network driver Kconfig start moving, with future categorizations as 
well -- the overall goal being that eventually you'll select among 
several classes of devices, instead of today's innumerably long list of 
devices from the first days of ethernet to today's 10Gbps.

The easy fix is just to cset -x the change I commited.  Ethernet and 
"gigabit ethernet" will be in separate menus, but there will not a 
separate CONFIG_NET_GIGE option.  I won't complain, if people feel 
'cset -x' is warranted.  It's just a config option, not anything 
terribly important[1]

	Jeff


[1] Of course, this is lkml, the big arguments are always over 
trivialities like this ;-)

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Christoph Hellwig]

On Sun, May 30, 2004 at 01:06:53PM -0700, Linus Torvalds wrote:
> is it really sane these days to split out gigabit from the "regular" 
> ethernets? gigabit ethernet is getting quite a bit more common than it 
> used to be, and a lot of the gigabit devices are just "standard ethernet" 
> as far as the user is concerned, and in fact they are often _used_ in just 
> regular 10/100Mbps setups.

Or have a gige chip that's crippled to do only 10/100, like the sungem
on my ibook.

Re: [patch] Re: Gigabit Kconfig problems with yesterday's update

[Paul Mackerras]

Christoph Hellwig writes:

> Or have a gige chip that's crippled to do only 10/100, like the sungem
> on my ibook.

I think it is not that the sungem itself is crippled, rather that the
phy can only do 10/100.  I agree that the division between 10/100 and
gigabit ethernet is pretty useless.  I note that sungem didn't get put
in the gigabit list, for instance.

Paul.