Skip to contentSkip to conversation

bcm43xx regression in 2.6.24 (with patch)

PATCH Alexey Zaytsev 10 participants 60 messages Feb 22, 2008 – Feb 29, 2008

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hello.

The bcm43xx driver won't work any more, if the b44 Ethernet driver is enabled. This happens because the b44 driver needlessly enables the b43_pci_bridge code, which claims the same pci ids as the bcm43xx driver. The b43_pci_bridge code is needed for the b43{legacy} drivers, but for the b44, only the "ssb pci core" is needed.

This patch separates the ssb b43 pci bridge and the ssb pci core config options and enables only the needed ones.

Greg, how do you think, should this do a -stable material? If the line number limit really matters, I could remove the comments in the b43{legacy} Kconfigs. ;)

PS:
Yes, the bcm43xx driver is deprecated, but the b43 driver did not work for me right out of the box (firmware updated). Please don't remove the bc43xx driver until there are really no problems with the new one.

I'll try the new driver once again this weekend, and will send a bug report when I'll be sure it is not a stupid mistake.

---

Use a separate config option for the b43 pci to ssb bridge.

The bridge code was unnecessary enabled by the b44 driver, but it prevents the bcm43xx driver from being loaded, as the bridge claims the same pci ids.

Now we enable the birdge only if the b43{legacy} drivers are selected.

---

drivers/net/wireless/b43/Kconfig       |    4 ++++
drivers/net/wireless/b43legacy/Kconfig |    4 ++++
drivers/net/wireless/bcm43xx/Kconfig   |    2 +-
drivers/ssb/Kconfig                    |    5 +++++
drivers/ssb/Makefile                   |    2 +-
drivers/ssb/ssb_private.h              |    4 ++--
6 files changed, 17 insertions(+), 4 deletions(-)
drivers/net/wireless/b43/Kconfig +4 -0
index fdbc351..8b7043a 100644
@@ -24,6 +24,9 @@ config B43
         This driver uses V4 firmware, which must be installed separately using
         b43-fwcutter.
 
+         Building this driver prevents the now deprecated bcm43xx driver from being
+         built.
+
         This driver can be built as a module (recommended) that will be called "b43".
         If unsure, say M.
 
@@ -32,6 +35,7 @@ config B43_PCI_AUTOSELECT
       bool
       depends on B43 && SSB_PCIHOST_POSSIBLE
       select SSB_PCIHOST
+       select SSB_B43_PCI_BRIDGE
       default y
 

# Auto-select SSB PCICORE driver, if possible

drivers/net/wireless/b43legacy/Kconfig +4 -0
index 7e23ec2..48f8e7d 100644
@@ -17,6 +17,9 @@ config B43LEGACY
         This driver uses V3 firmware, which must be installed separately using
         b43-fwcutter.
 
+         Building this driver prevents the now deprecated bcm43xx driver from being
+         built.
+
         This driver can be built as a module (recommended) that will be
         called "b43legacy". If unsure, say M.
 
@@ -25,6 +28,7 @@ config B43LEGACY_PCI_AUTOSELECT
       bool
       depends on B43LEGACY && SSB_PCIHOST_POSSIBLE
       select SSB_PCIHOST
+       select SSB_B43_PCI_BRIDGE
       default y
 

# Auto-select SSB PCICORE driver, if possible

drivers/net/wireless/bcm43xx/Kconfig +0 -0
index ce397e4..32068d8 100644
@@ -1,6 +1,6 @@

config BCM43XX
tristate "Broadcom BCM43xx wireless support"
- depends on PCI && IEEE80211 && IEEE80211_SOFTMAC && WLAN_80211 && EXPERIMENTAL
+ depends on PCI && IEEE80211 && IEEE80211_SOFTMAC && WLAN_80211 && !SSB_B43_PCI_BRIDGE && EXPERIMENTAL
select WIRELESS_EXT
select FW_LOADER
select HW_RANDOM

drivers/ssb/Kconfig +5 -0
index d976660..6f627f3 100644
@@ -35,6 +35,11 @@ config SSB_PCIHOST
 
         If unsure, say Y
 
+config SSB_B43_PCI_BRIDGE
+       bool
+       depends on SSB_PCIHOST
+       default n
+

config SSB_PCMCIAHOST_POSSIBLE

       bool
       depends on SSB && (PCMCIA = y || PCMCIA = SSB) && EXPERIMENTAL
drivers/ssb/Makefile +0 -0
index 7be3975..1bf35f0 100644
@@ -13,6 +13,6 @@ ssb-$(CONFIG_SSB_DRIVER_PCICORE)      += driver_pcicore.o
 
# b43 pci-ssb-bridge driver
# Not strictly a part of SSB, but kept here for convenience
-ssb-$(CONFIG_SSB_PCIHOST)              += b43_pci_bridge.o
+ssb-$(CONFIG_SSB_B43_PCI_BRIDGE)       += b43_pci_bridge.o

obj-$(CONFIG_SSB) += ssb.o

drivers/ssb/ssb_private.h +0 -0
index a789364..21eca2b 100644
@@ -120,10 +120,10 @@ extern int ssb_devices_thaw(struct ssb_bus *bus);

extern struct ssb_bus *ssb_pci_dev_to_bus(struct pci_dev *pdev);

/* b43_pci_bridge.c */
-#ifdef CONFIG_SSB_PCIHOST
+#ifdef CONFIG_SSB_B43_PCI_BRIDGE
extern int __init b43_pci_ssb_bridge_init(void);
extern void __exit b43_pci_ssb_bridge_exit(void);
-#else /* CONFIG_SSB_PCIHOST */
+#else /* CONFIG_SSB_B43_PCI_BRIDGR */
static inline int b43_pci_ssb_bridge_init(void)

{
       return 0;
#1 lore raw
Quoted 10 lines · On Friday 22 February 2008 12:17:15 Alexey Zaytsev wrote:

Hello.

The bcm43xx driver won't work any more, if the b44 Ethernet driver is enabled. This happens because the b44 driver needlessly enables the b43_pci_bridge code, which claims the same pci ids as the bcm43xx driver. The b43_pci_bridge code is needed for the b43{legacy} drivers, but for the b44, only the "ssb pci core" is needed.

Quoted 2 lines

This patch separates the ssb b43 pci bridge and the ssb pci core config options and enables only the needed ones.

Nack. Switch to b43. bcm43xx is going to be removed anyway. I'm not going to play these kconfig SELECT tricks anymore. We had _lots_ of bugs there. People submitted patches that obviously looked OK and still they turned out to break some dependencies. KConfig SELECT is a feature from hell.

Signature · 1 line
Greetings Michael.
#2 lore raw
Quoted 10 lines · Michael Buesch
Quoted 9 lines · On Friday 22 February 2008 12:17:15 Alexey Zaytsev wrote:

Hello.

The bcm43xx driver won't work any more, if the b44 Ethernet driver is enabled. This happens because the b44 driver needlessly enables the b43_pci_bridge code, which claims the same pci ids as the bcm43xx driver. The b43_pci_bridge code is needed for the b43{legacy} drivers, but for the b44, only the "ssb pci core" is needed.

Quoted 9 lines
Quoted 2 lines

This patch separates the ssb b43 pci bridge and the ssb pci core config options and enables only the needed ones.

Nack. Switch to b43. bcm43xx is going to be removed anyway. I'm not going to play these kconfig SELECT tricks anymore. We had _lots_ of bugs there. People submitted patches that obviously looked OK and still they turned out to break some dependencies. KConfig SELECT is a feature from hell.

Sorry, I don't get it. You are going to remove the (somehow) working driver, while there are known problems with the new one? Why? To me it sounds like _breaking old but working code_ to get _more bug reports on the new code_. But, what was the goal providing users with working drivers, or getting more bug reports?

And please look at the problem from a user's view point. I'm happily using the 2.6.23 kernel. The bcm43xx driver is the
only one available. And it works somehow (well, no s2ram).
Now I upgrade to the 2.6.24 kernel. I see there is a new b43 driver and try it. For some reason it does not work, even after I follow the
firmware upgrading instructions. Not a big deal, I clearly understand that you have to work with reverse-engineered
specs and even with good specs, bugs happen, no problem.
I'll try to debug it in the weekend and maybe will send a
bug report or patch.

But right now I have other business, sorry. The problem is, now also the old (but not now marked as deprecated) driver stops working.

If this is not a repgession, than I don't know what is. And if it is a regression, it should be fixed at least in the 2.6.24.y series, do you agree?

I have provided a patch that I believe is trivial, that I have tested with all possible config option combinations I thought were possible, and that fixes the regression. If you have a reason to believe it is wrong, please say it, I won't be offended. If there is a problem with the patch, I'll gladly fix and resend it.

This Nack leaves me with one option - to switch back to 2.6.23. Sorry, no testing, probably more bugs not found in 2.6.24. ;(

#3 lore raw
Quoted 3 lines · On Friday 22 February 2008 19:10:39 Alexey Zaytsev wrote:

Sorry, I don't get it. You are going to remove the (somehow) working driver, while there are known problems with the new one? Why?

Because the new driver works, if you just set it up right. Until now every "breakage" was just a usage error.

Quoted 3 lines

And please look at the problem from a user's view point. I'm happily using the 2.6.23 kernel. The bcm43xx driver is the only one available. And it works somehow (well, no s2ram).

b43 supports s2ram and s2disk.

Quoted 3 lines

If this is not a repgession, than I don't know what is. And if it is a regression, it should be fixed at least in the 2.6.24.y series, do you agree?

No. Playing with kconfig SELECT is really _nothing_ for a -stable series. I am _not_ going to be responsible for any
breakages. And using SELECT _does_ break things. We had dozens of bugs until all the damn dependencies were right. Even "Obviously Correct (tm)" changes to SELECT _do_ break on weird configurations.
See the bcm43xx mail archives for examples.

If you need that patch for some reason, please apply it locally. The problem will automatically get fixed when bcm43xx goes away, soon.

Quoted 9 lines

I have provided a patch that I believe is trivial, that I have tested with all possible config option combinations I thought were possible, and that fixes the regression. If you have a reason to believe it is wrong, please say it, I won't be offended. If there is a problem with the patch, I'll gladly fix and resend it.

This Nack leaves me with one option - to switch back to 2.6.23. Sorry, no testing, probably more bugs not found in 2.6.24. ;(

It is not my problem, if you refuse to use b43.
You also still refuse to tell me details about your card and _what_ does not work. I do own lots of different card and they
all work fine with b43. There's one exception, the 4311 rev 3 (or something, don't quite remember). But patches are available and will ship in 2.6.25. bcm43xx won't get removed until that shipped.

Signature · 1 line
Greetings Michael.
#57 lore raw
In reply to Michael Buesch
Quoted 10 lines · Michael Buesch
Quoted 9 lines · On Friday 22 February 2008 12:17:15 Alexey Zaytsev wrote:

Hello.

The bcm43xx driver won't work any more, if the b44 Ethernet driver is enabled. This happens because the b44 driver needlessly enables the b43_pci_bridge code, which claims the same pci ids as the bcm43xx driver. The b43_pci_bridge code is needed for the b43{legacy} drivers, but for the b44, only the "ssb pci core" is needed.

Quoted 4 lines
Quoted 2 lines

This patch separates the ssb b43 pci bridge and the ssb pci core config options and enables only the needed ones.

Nack. Switch to b43. bcm43xx is going to be removed anyway.

Since when this is a valid reason to NACK a patch which fixes a regression in the 'stable' kernel ?

Probably you may want to ask first whatever 'b43' works for Alexey ? and most probably the answer will be 'no' because that is the only reason one would want to fix the *old* one when *both* drivers are enabled.

Really 'Switch to foo' and 'foo_old will be removed sometimes' is not a 'valid' reason to NACK anything as long you 'have foo and foo_old' *supported* in that kernel.

Quoted 1 line

I'm not going to play these kconfig SELECT tricks anymore.

Fix it different then.

Quoted 4 lines

We had _lots_ of bugs there. People submitted patches that obviously looked OK and still they turned out to break some dependencies. KConfig SELECT is a feature from hell.

Yes it is but that is still not a valid reason to NACK that patch , I'm sorry.

Regards,

Gabriel

#58 lore raw
Quoted 3 lines · On Friday 22 February 2008 18:51:32 Gabriel C wrote:
Quoted 1 line

I'm not going to play these kconfig SELECT tricks anymore.

Fix it different then.

Please do so.

Quoted 1 line

Yes it is but that is still not a valid reason to NACK that patch , I'm sorry.

NACK means I (being the maintainer of the modified code) won't sign-off this patch. Nothing more or less. If some upstream maintainer decides to bypass me and apply the patch nevertheless is not of my business. But I will not sign-off or ack this patch.

Signature · 1 line
Greetings Michael.
#4 lore raw
In reply to Michael Buesch

[replying from my non-work account]

Quoted 20 lines · Michael Buesch
Quoted 3 lines · On Friday 22 February 2008 19:10:39 Alexey Zaytsev wrote:

Sorry, I don't get it. You are going to remove the (somehow) working driver, while there are known problems with the new one? Why?

Because the new driver works, if you just set it up right. Until now every "breakage" was just a usage error.

Quoted 3 lines

And please look at the problem from a user's view point. I'm happily using the 2.6.23 kernel. The bcm43xx driver is the only one available. And it works somehow (well, no s2ram).

b43 supports s2ram and s2disk.

Quoted 3 lines

If this is not a repgession, than I don't know what is. And if it is a regression, it should be fixed at least in the 2.6.24.y series, do you agree?

Quoted 7 lines

No. Playing with kconfig SELECT is really _nothing_ for a -stable series. I am _not_ going to be responsible for any
breakages. And using SELECT _does_ break things. We had dozens of bugs until all the damn dependencies were right. Even "Obviously Correct (tm)" changes to SELECT _do_ break on weird configurations.
See the bcm43xx mail archives for examples.

Well, someone obviously played with the SELECTs, and broke the bcm43xx driver. Now I've got a patch that I believe gets the SELECTs right and fixes the driver, so
what is wrong?

Quoted 14 lines

If you need that patch for some reason, please apply it locally. The problem will automatically get fixed when bcm43xx goes away, soon.

Quoted 9 lines

I have provided a patch that I believe is trivial, that I have tested with all possible config option combinations I thought were possible, and that fixes the regression. If you have a reason to believe it is wrong, please say it, I won't be offended. If there is a problem with the patch, I'll gladly fix and resend it.

This Nack leaves me with one option - to switch back to 2.6.23. Sorry, no testing, probably more bugs not found in 2.6.24. ;(

Quoted 6 lines

It is not my problem, if you refuse to use b43.
You also still refuse to tell me details about your card and _what_ does not work. I do own lots of different card and they
all work fine with b43. There's one exception, the 4311 rev 3 (or something, don't quite remember). But patches are available and will ship in 2.6.25. bcm43xx won't get removed until that shipped.

Yes, it's a 4311 rev 01, but I'm probably was just too lame to upgrade the firmware or something. :E

I really don't get it, what is going on here? You state that the new b32 driver has problems on some hardware, where the old bcm43xx driver just works. And at the same time, you are surprised that I "refuse" to use the b43 driver and push patches for the bcm43xx driver you broke... Oh, really, why?!

#5 lore raw
Quoted 7 lines · On Friday 22 February 2008 21:06:00 Alexey Zaytsev wrote:
Quoted 6 lines

It is not my problem, if you refuse to use b43.
You also still refuse to tell me details about your card and _what_ does not work. I do own lots of different card and they
all work fine with b43. There's one exception, the 4311 rev 3 (or something, don't quite remember). But patches are available and will ship in 2.6.25. bcm43xx won't get removed until that shipped.

Quoted 7 lines

Yes, it's a 4311 rev 01, but I'm probably was just too lame to upgrade the firmware or something. :E

I really don't get it, what is going on here? You state that the new b32 driver has problems on some hardware, where the old bcm43xx driver just works. And at the same time, you are surprised that I "refuse" to use the b43 driver and push patches for the bcm43xx driver you broke... Oh, really, why?!

So, please find someone who will sign-off your patch. I won't. What's so hard to understand about that? Do I _have_ to sign off all patches random people send to me?
I do _not_ want to be made responsible for that patch by signing it off. It is as simple as that.
And I officially do not care about bcm43xx since a year and a half anymore. So why should I ACK it or sign it off?

Signature · 1 line
Greetings Michael.
#6 lore raw
Quoted 16 lines · Michael Buesch
Quoted 7 lines · On Friday 22 February 2008 21:06:00 Alexey Zaytsev wrote:
Quoted 6 lines

It is not my problem, if you refuse to use b43.
You also still refuse to tell me details about your card and _what_ does not work. I do own lots of different card and they
all work fine with b43. There's one exception, the 4311 rev 3 (or something, don't quite remember). But patches are available and will ship in 2.6.25. bcm43xx won't get removed until that shipped.

Quoted 7 lines

Yes, it's a 4311 rev 01, but I'm probably was just too lame to upgrade the firmware or something. :E

I really don't get it, what is going on here? You state that the new b32 driver has problems on some hardware, where the old bcm43xx driver just works. And at the same time, you are surprised that I "refuse" to use the b43 driver and push patches for the bcm43xx driver you broke... Oh, really, why?!

Quoted 8 lines

So, please find someone who will sign-off your patch. I won't. What's so hard to understand about that? Do I _have_ to sign off all patches random people send to me?
I do _not_ want to be made responsible for that patch by signing it off. It is as simple as that.
And I officially do not care about bcm43xx since a year and a half anymore. So why should I ACK it or sign it off?

I thought that there was a rule that if you break something in the kernel, you normally would be the one who fixes things up. Sorry, it looks I was wrong.

I'll resend the patch directly to Greg KH and Jeff Garzik for -stable and 2.6.25 inclusion.

#59 lore raw
In reply to Michael Buesch
Quoted 5 lines · Michael Buesch
Quoted 3 lines · On Friday 22 February 2008 18:51:32 Gabriel C wrote:
Quoted 1 line

I'm not going to play these kconfig SELECT tricks anymore.

Fix it different then.

Quoted 8 lines

Please do so.

Quoted 1 line

Yes it is but that is still not a valid reason to NACK that patch , I'm sorry.

NACK means I (being the maintainer of the modified code) won't sign-off this patch. Nothing more or less. If some upstream maintainer decides to bypass me and apply the patch nevertheless is not of my business. But I will not sign-off or ack this patch.

As the maintainer will not accept this, I'm not going to take it for the -stable tree either, sorry.

thanks,

greg k-h

#7 lore raw
In reply to Alexey Zaytsev
Quoted 26 lines · Alexey Zaytsev
Quoted 16 lines · Michael Buesch
Quoted 7 lines · On Friday 22 February 2008 21:06:00 Alexey Zaytsev wrote:
Quoted 6 lines

It is not my problem, if you refuse to use b43.
You also still refuse to tell me details about your card and _what_ does not work. I do own lots of different card and they
all work fine with b43. There's one exception, the 4311 rev 3 (or something, don't quite remember). But patches are available and will ship in 2.6.25. bcm43xx won't get removed until that shipped.

Quoted 7 lines

Yes, it's a 4311 rev 01, but I'm probably was just too lame to upgrade the firmware or something. :E

I really don't get it, what is going on here? You state that the new b32 driver has problems on some hardware, where the old bcm43xx driver just works. And at the same time, you are surprised that I "refuse" to use the b43 driver and push patches for the bcm43xx driver you broke... Oh, really, why?!

Quoted 8 lines

So, please find someone who will sign-off your patch. I won't. What's so hard to understand about that? Do I _have_ to sign off all patches random people send to me?
I do _not_ want to be made responsible for that patch by signing it off. It is as simple as that.
And I officially do not care about bcm43xx since a year and a half anymore. So why should I ACK it or sign it off?

Quoted 5 lines

I thought that there was a rule that if you break something in the kernel, you normally would be the one who fixes things up. Sorry, it looks I was wrong.

I'll resend the patch directly to Greg KH and Jeff Garzik for -stable and 2.6.25 inclusion.

One of the rules for the -stable tree is that the patch is accepted by the maintainer and it is already in the upstream kernel version. Both of these requirements do not seem to be met here.

I suggest you work _with_ Michael instead of against him to try to determine why your device doesn't work properly with the recommended driver.

thanks,

greg k-h

#8 lore raw
Quoted 31 lines · Greg KH
Quoted 29 lines · Alexey Zaytsev
Quoted 16 lines · Michael Buesch
Quoted 7 lines · On Friday 22 February 2008 21:06:00 Alexey Zaytsev wrote:
Quoted 6 lines

It is not my problem, if you refuse to use b43.
You also still refuse to tell me details about your card and _what_ does not work. I do own lots of different card and they
all work fine with b43. There's one exception, the 4311 rev 3 (or something, don't quite remember). But patches are available and will ship in 2.6.25. bcm43xx won't get removed until that shipped.

Quoted 7 lines

Yes, it's a 4311 rev 01, but I'm probably was just too lame to upgrade the firmware or something. :E

I really don't get it, what is going on here? You state that the new b32 driver has problems on some hardware, where the old bcm43xx driver just works. And at the same time, you are surprised that I "refuse" to use the b43 driver and push patches for the bcm43xx driver you broke... Oh, really, why?!

Quoted 8 lines

So, please find someone who will sign-off your patch. I won't. What's so hard to understand about that? Do I _have_ to sign off all patches random people send to me?
I do _not_ want to be made responsible for that patch by signing it off. It is as simple as that.
And I officially do not care about bcm43xx since a year and a half anymore. So why should I ACK it or sign it off?

I thought that there was a rule that if you break something in the kernel, you normally would be the one who fixes things up. Sorry, it looks I was wrong.

Quoted 6 lines
Quoted 2 lines

I'll resend the patch directly to Greg KH and Jeff Garzik for -stable and 2.6.25 inclusion.

One of the rules for the -stable tree is that the patch is accepted by the maintainer and it is already in the upstream kernel version. Both of these requirements do not seem to be met here.

Well, it looks like Michael is not the bcm43xx maintaner. I sent the patch to him,
because it was his code that broke the driver, and I thought it would be easy for him to review my patch, as it touches his code.

I'll resend the patch to the proper maintainers now.

Quoted 4 lines

I suggest you work _with_ Michael instead of against him to try to determine why your device doesn't work properly with the recommended driver.

Well, if you read my first email, that is exactly what I intended to do, but even if
Michael would be able to fix the b43 driver to work with my hardware, the code will only show up in 2.6.25, leaving some 2.6.24 users with broken wifi. So I thought
it would be a good thing to add my simple fix that enabled the old driver to the -stable tree, so that we could have working wifi soon.

This is still my intension, I'll resend to the proper maintainters.

Quoted 5 lines

thanks,

greg k-h

#9 lore raw
Quoted 4 lines · Alexey Zaytsev

Well, it looks like Michael is not the bcm43xx maintaner. I sent the patch to him,
because it was his code that broke the driver, and I thought it would be easy for him to review my patch, as it touches his code.

See? I'm tired of this "how dare can you break my kernel!?" bullshit. That's exactly the reason why I NACKed this patch.
I do _not_ understand the KConfig SELECT logic. And I do think almost nobody does understand how that all works together.
In the past people came with similiar patches like yours that looked obviously OK. They said sentences like "it is trivial to get the SELECT logics right". But it turned out they were wrong and it introduced other regressions that I was made responsible for.

SELECT is _extremely_ difficult to get right, as it completely ignores dependencies. See all this FOOBAR_POSSIBLE select logic that we use in SSB to get SELECT working correctly with dependencies.

So my solution for this particular breakage you are seeing here is to wait for the bcm43xx removal, which will happen soon. That will fix it and will have almost zero chance to introduce new bugs.

Quoted 5 lines

Well, if you read my first email, that is exactly what I intended to do, but even if
Michael would be able to fix the b43 driver to work with my hardware, the code will only show up in 2.6.25, leaving some 2.6.24 users with broken wifi. So I thought

Blah. The people with a bcm4311 revision 1 wireless card plus a bcm44xx ethernet card. You can count those people on two fingers.

Quoted 4 lines

it would be a good thing to add my simple fix that enabled the old driver to the -stable tree, so that we could have working wifi soon.

This is still my intension, I'll resend to the proper maintainters.

Ok thanks. We'll see if it really was a simple fix then.
If it turns out to break something you will get mail. :)
You can send this patch to a netdev maintainer or the wireless maintainer. Maybe one of those will sign it off. Good luck.

#10 lore raw
In reply to Michael Buesch
Quoted 6 lines · Michael Buesch
Quoted 3 lines

If this is not a repgession, than I don't know what is. And if it is a regression, it should be fixed at least in the 2.6.24.y series, do you agree?

No. Playing with kconfig SELECT is really _nothing_ for a -stable series. I am _not_ going to be responsible for any breakages. [...]

well, i've reviewed this thread and it's pretty apparent to any outside observer that you as a maintainer are ignoring Alexey Zaytsev's pretty reasonable request for a fix.

Alexey had a problem, he analyzed it, he found a fix which he tested, and he even has offered to test anything you send his way:

|| I have provided a patch that I believe is trivial, that I have tested 
|| with all possible config option combinations I thought were possible, 
|| and that fixes the regression. If you have a reason to believe it is 
|| wrong, please say it, I won't be offended. If there is a problem with 
|| the patch, I'll gladly fix and resend it.

that's about the most friendly tester attitude that is imaginable.

but what were you able to make out of that positive attitude? The only things i've seen you send his way were insults and general handwaving about how his patch breaks stuff (without providing a _shred_ of evidence).

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to _your own buggy driver code_ without providing an alternative fix. Kudos.

Ingo

#11 lore raw
Quoted 6 lines · Ingo Molnar

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to _your own buggy driver code_ without providing an alternative fix. Kudos.

Agreed. Alexey, did you identify a specific git commit that caused the regression? Can we just revert that from 2.6.24? Michael, even if _you're_ planning to remove bcm43xx we must not let it regress until it's gone.

#12 lore raw
Quoted 4 lines · Pekka Enberg

Agreed. Alexey, did you identify a specific git commit that caused the regression? Can we just revert that from 2.6.24? Michael, even if _you're_ planning to remove bcm43xx we must not let it regress until it's gone.

btw., if the best answer is: "do not enable b43 and bcm43xx at once in the .config because we messed up the dependencies and fixing it in 2.6.24-stable is unfortunately too risky", then that's a perfectly fine answer for -stable. Then the fix would be to exclude each other in the Kconfig space (which is not risky). But just ignoring bugreports and NAK-ing a fix is not a valid answer.

Ingo

#13 lore raw
Quoted 8 lines · Ingo Molnar
Quoted 4 lines · Pekka Enberg

Agreed. Alexey, did you identify a specific git commit that caused the regression? Can we just revert that from 2.6.24? Michael, even if _you're_ planning to remove bcm43xx we must not let it regress until it's gone.

Quoted 7 lines

btw., if the best answer is: "do not enable b43 and bcm43xx at once in the .config because we messed up the dependencies and fixing it in 2.6.24-stable is unfortunately too risky", then that's a perfectly fine answer for -stable. Then the fix would be to exclude each other in the Kconfig space (which is not risky). But just ignoring bugreports and NAK-ing a fix is not a valid answer.

The problem is not with enabling both b43 and bcm43xx (will, whis won't work anyway, and there is no chance fixing it). The problem is with enabling the bcm43xx wifi driver and the b44 Ethernet driver. The ethernet driver then wrongly autosolects some config options that prevent the bcm43xx driver from loading, so I'm left with either no wifi or no ethenet.

Quoted 9 lines

Ingo

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#14 lore raw
Quoted 5 lines · Alexey Zaytsev

The problem is not with enabling both b43 and bcm43xx (will, whis won't work anyway, and there is no chance fixing it). The problem is with enabling the bcm43xx wifi driver and the b44 Ethernet driver. The ethernet driver then wrongly autosolects some config options that prevent the bcm43xx driver from loading, so I'm left with either no wifi or no ethenet.

So does reverting commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") fix the regression?

#15 lore raw
Quoted 11 lines · Pekka Enberg
Quoted 5 lines · Alexey Zaytsev

The problem is not with enabling both b43 and bcm43xx (will, whis won't work anyway, and there is no chance fixing it). The problem is with enabling the bcm43xx wifi driver and the b44 Ethernet driver. The ethernet driver then wrongly autosolects some config options that prevent the bcm43xx driver from loading, so I'm left with either no wifi or no ethenet.

So does reverting commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") fix the regression?

Compiling it right now.

I'm sure it fixes the problem, but I don't feel like we should revert it. I can't
be sure it won't cause any other problems, and anyway, this does not look right a solution. Any other driver (rightfully) requiring the SSB_HOST would then also break the bcm43xx driver.

I'll send an simplified patch in a few minutes.

#16 lore raw

Hi Alexey,

Quoted 3 lines · Pekka Enberg
Quoted 3 lines

So does reverting commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") fix the regression?

Quoted 9 lines · Alexey Zaytsev

Compiling it right now.

I'm sure it fixes the problem, but I don't feel like we should revert it. I can't
be sure it won't cause any other problems, and anyway, this does not look right a solution. Any other driver (rightfully) requiring the SSB_HOST would then also break the bcm43xx driver.

I'll send an simplified patch in a few minutes.

The point here is that it's Michael's patch and I don't want to start blaming anyone until you've verified it actually broke your setup. And if the problem was actually caused by that patch, Michael needs to either ACK your patch or fix it himself.

Pekka

#17 lore raw
Quoted 19 lines · Pekka J Enberg

Hi Alexey,

Quoted 3 lines · Pekka Enberg
Quoted 3 lines

So does reverting commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") fix the regression?

Quoted 8 lines · Alexey Zaytsev

Compiling it right now.

I'm sure it fixes the problem, but I don't feel like we should revert it. I can't
be sure it won't cause any other problems, and anyway, this does not look right a solution. Any other driver (rightfully) requiring the SSB_HOST would then also break the bcm43xx driver.

Quoted 6 lines
Quoted 1 line

I'll send an simplified patch in a few minutes.

The point here is that it's Michael's patch and I don't want to start blaming anyone until you've verified it actually broke your setup. And if the problem was actually caused by that patch, Michael needs to either ACK your patch or fix it himself.

Yes, the problem was caused by this patch. With it reverted, I've got both ethernet
and wifi working.

Quoted 3 lines

Pekka

#18 lore raw

Hi,

Quoted 2 lines · Pekka Enberg
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

So does reverting commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") fix the regression?

Quoted 2 lines · Alexey Zaytsev

Yes, the problem was caused by this patch. With it reverted, I've got both ethernet and wifi working.

Michael, I am always glad to help out but I would much appreciate if you could show some effort on debugging your own bugs in the future. Perhaps you could now work with Alexey to fix it?

#34 lore raw
In reply to Ingo Molnar
Quoted 23 lines · On Saturday 23 February 2008 12:07:51 Ingo Molnar wrote:
Quoted 6 lines · Michael Buesch
Quoted 3 lines

If this is not a repgession, than I don't know what is. And if it is a regression, it should be fixed at least in the 2.6.24.y series, do you agree?

No. Playing with kconfig SELECT is really _nothing_ for a -stable series. I am _not_ going to be responsible for any breakages. [...]

well, i've reviewed this thread and it's pretty apparent to any outside observer that you as a maintainer are ignoring Alexey Zaytsev's pretty reasonable request for a fix.

Alexey had a problem, he analyzed it, he found a fix which he tested, and he even has offered to test anything you send his way:

|| I have provided a patch that I believe is trivial, that I have tested 
|| with all possible config option combinations I thought were possible, 
|| and that fixes the regression. If you have a reason to believe it is 
|| wrong, please say it, I won't be offended. If there is a problem with 
|| the patch, I'll gladly fix and resend it.
Quoted 6 lines

that's about the most friendly tester attitude that is imaginable.

but what were you able to make out of that positive attitude? The only things i've seen you send his way were insults and general handwaving about how his patch breaks stuff (without providing a _shred_ of evidence).

blah:

Quoted 6 lines

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to _your own buggy driver code_ without providing an alternative fix. Kudos.

So I am forced to sign-off random patches people send to me? I explained why I do not. If you do not like that, please do sign it off. If you do think the patch is correct, please _do_ sign it off Ingo.

This problem will fix itself by switching to b43 and dropping bcm43xx. _That_ is my way to fix the bug.
I don't understand all the SELECT implications, so I'm not going to introduce more of them. Because if the next regression appears from I SELECT that I signed off
goto blah;

Signature · 1 line
Greetings Michael.
#35 lore raw
In reply to Ingo Molnar
Quoted 4 lines · On Saturday 23 February 2008 12:07:51 Ingo Molnar wrote:

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to

The insults being? A few quotes, please.

Quoted 2 lines

_your own buggy driver code_ without providing an alternative fix. Kudos.

How dare can you break my kernel, yeah... . We know that. That my patches _fixed_ the kernel for thousands of people doesn't matter. What matters is the one person for whom it broke. So let's apply a questionable fix for it that we don't understand. Hopefully that fixes it and doesn't break it again for somebody else... .

I provided an alternative fix. In my very first email. Ingo, if you know that this fix is right, then please sign it off so we can apply it.

Signature · 1 line
Greetings Michael.
#36 lore raw
Quoted 7 lines · Michael Buesch
Quoted 4 lines · On Saturday 23 February 2008 12:07:51 Ingo Molnar wrote:

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to

The insults being? A few quotes, please.

" I'm tired of this "how dare can you break my kernel!?" bullshit. "
...
" Blah. The people with a bcm4311 revision 1 wireless card plus a
bcm44xx ethernet card. You can count those people on two fingers. "
...

    http://lkml.org/lkml/2008/2/23/3

	Ingo
#37 lore raw
Quoted 11 lines · On Saturday 23 February 2008 17:44:33 Ingo Molnar wrote:
Quoted 7 lines · Michael Buesch
Quoted 4 lines · On Saturday 23 February 2008 12:07:51 Ingo Molnar wrote:

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to

The insults being? A few quotes, please.

Quoted 7 lines

" I'm tired of this "how dare can you break my kernel!?" bullshit. "
...
" Blah. The people with a bcm4311 revision 1 wireless card plus a
bcm44xx ethernet card. You can count those people on two fingers. "
...

    http://lkml.org/lkml/2008/2/23/3

Hm, maybe my fault and I don't know the english language enough, but these do not count as personal insults here where I live in Germany. Especially the second one is in no way an insult here. It merely tells you that the number of people having this issue is very low.

I'm sorry if that sounded like insults. That wasn't my intention.

Signature · 1 line
Greetings Michael.
#38 lore raw
In reply to Michael Buesch
Quoted 9 lines · Michael Buesch
Quoted 4 lines · On Saturday 23 February 2008 12:07:51 Ingo Molnar wrote:

I have to say, after having observed multiple incidents around b43 in the past few months you are one of the worst driver maintainers i've ever seen on lkml: you are ignoring regressions, you are frequently insulting our testers and now you even have the gall to NAK a patch to

The insults being? A few quotes, please.

If you really want to know, the
"Because the new driver works, if you just set it up right." for me was clearly a hint that I'm just an other imcompetent user, who can't even follow the instructions from linuxwireless.org. And you knew that the new driver did no work with the bcm4311 chips, which is the sad thing.

Quoted 7 lines
Quoted 3 lines

_your own buggy driver code_ without providing an alternative fix. Kudos.

How dare can you break my kernel, yeah... . We know that. That my patches _fixed_ the kernel for thousands of people doesn't matter.

Not true at all. I my first and second emails, there was not a single word about who broke the driver. There were only the fix, and the explanation, why the fix is needed. I really appreciate the work done on the b43 driver, and I'd emmediately switch to it, if only it worked for me*.

Quoted 1 line

What matters is the one person for whom it broke.

I'm the one person who investigated the problem, wrote a patch and sent it to the lkml. After lurking in #bcm-users for a day, I saw quite a few users asking why the b43 driver did not work for them, and the best answer they got, was got read the linuxwireless.org. Are you sure they all were just unable to upgrade the firmware?

And by the way, the bcm4311 and b44 combination is not something I assembled just to piss you off. It's what HP puts in their HP Compaq nx7300 laptops. They were like the cheapest Core Duo laptops a year ago, and I'm sure they were not exclusively made.

Quoted 5 lines

So let's
apply a questionable fix for it that we don't understand. Hopefully that fixes it and doesn't break it again for somebody else... .

I provided an alternative fix. In my very first email.

Sorry, your fix does not work for my hardware. And you know it.

Quoted 13 lines

Ingo, if you know that this fix is right, then please sign it off so we can apply it.

--
Greetings Michael.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#39 lore raw
Quoted 5 lines · On Saturday 23 February 2008 22:32:46 Alexey Zaytsev wrote:
Quoted 1 line

The insults being? A few quotes, please.

If you really want to know, the
"Because the new driver works, if you just set it up right." for me was clearly a hint that I'm just an other imcompetent user, who can't even follow the instructions from linuxwireless.org.

Well. That's 99% of all bugreports I receive. You will understand that I won't explain the same crap that's written on linuxwireless.org over and over 20 times each day to random people.
And yes, you belong to the group of my "random people", as I don't remember your name from the top of my head.
So in 99% of the cases the user did something wrong.

But I still don't see the insult.

Quoted 2 lines

And you knew that the new driver did no work with the bcm4311 chips, which is the sad thing.

That is not true. It doesn't work with exactly _one_ revision of the bcm4311 card. And that is already fixed in 2.6.25. I'd like to have that in .24-stable, too, but I guess it's too big. It changes some parts of the DMA engine code.

Quoted 4 lines

Not true at all. I my first and second emails, there was not a single word about who broke the driver. There were only the fix, and the explanation, why the fix is needed. I really appreciate the work done on the b43 driver, and I'd emmediately switch to it, if only it worked for me*.

Yeah. But I don't like that patch. Then people came and said stuff like I _have_ to accept it because it fixes stuff I broke. Excuse me, but nobody can force me to sign off a patch that I am not completely convinced of. _Even_ if it turns out that a commit made by me broke something. Especially, if there's already another fix for the situation queued up in the development kernel.

Quoted 6 lines

I'm the one person who investigated the problem, wrote a patch and sent it to the lkml. After lurking in #bcm-users for a day, I saw quite a few users asking why the b43 driver did not work for them, and the best answer they got, was got read the linuxwireless.org. Are you sure they all were just unable to upgrade the firmware?

Yeah. Experience shows that. Read the archives.

Quoted 5 lines

And by the way, the bcm4311 and b44 combination is not something I assembled just to piss you off. It's what HP puts in their HP Compaq nx7300 laptops. They were like the cheapest Core Duo laptops a year ago, and I'm sure they were not exclusively made.

So be it. And you guess what? We already sent a fix for those people upstream a long time ago. It is in 2.6.25.

Quoted 7 lines
Quoted 5 lines

So let's
apply a questionable fix for it that we don't understand. Hopefully that fixes it and doesn't break it again for somebody else... .

I provided an alternative fix. In my very first email.

Sorry, your fix does not work for my hardware. And you know it.

Did you _ever_ even try it?
http://linuxwireless.org/en/users/Download

Signature · 1 line
Greetings Michael.
#19 lore raw
In reply to Alexey Zaytsev
Quoted 5 lines · On Saturday 23 February 2008 12:32:50 Alexey Zaytsev wrote:

The problem is not with enabling both b43 and bcm43xx (will, whis won't work anyway, and there is no chance fixing it). The problem is with enabling the bcm43xx wifi driver and the b44 Ethernet driver. The ethernet driver then wrongly autosolects some config options that prevent the bcm43xx driver from loading, so I'm left with either no wifi or no ethenet.

Ok, so I just tested this on a machine and it is _not_ true. It works just fine to run b44/ssb with bcm43xx in parallel.

Can you please stop fooling me and spreading this FUD?

Signature · 1 line
Greetings Michael.
#20 lore raw
Quoted 7 lines · Michael Buesch
Quoted 5 lines · On Saturday 23 February 2008 12:32:50 Alexey Zaytsev wrote:

The problem is not with enabling both b43 and bcm43xx (will, whis won't work anyway, and there is no chance fixing it). The problem is with enabling the bcm43xx wifi driver and the b44 Ethernet driver. The ethernet driver then wrongly autosolects some config options that prevent the bcm43xx driver from loading, so I'm left with either no wifi or no ethenet.

Quoted 5 lines

Ok, so I just tested this on a machine and it is _not_ true. It works just fine to run b44/ssb with bcm43xx in parallel.

Can you please stop fooling me and spreading this FUD?

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right? Or the b43xx driver was built statically and the b44 and the ssb as modules.

Looks like this, kind of, ruins my statement that is is impossible to get working wifi in 2.6.24. Sorry, it was not intensinal. ;)

But there still are lots of valid combinations where the b43xx driver silently won't work. E.g. when either the ssb or the b44 modules are loaded prior to it, or when either of them are built statically (building the b43xx driver staticaly won't help here). And my patch fixes them.

So, my statement goes, the bug stays. Agreed?

#21 lore raw
Quoted 11 lines · Alexey Zaytsev
Quoted 10 lines · Michael Buesch
Quoted 5 lines · On Saturday 23 February 2008 12:32:50 Alexey Zaytsev wrote:

The problem is not with enabling both b43 and bcm43xx (will, whis won't work anyway, and there is no chance fixing it). The problem is with enabling the bcm43xx wifi driver and the b44 Ethernet driver. The ethernet driver then wrongly autosolects some config options that prevent the bcm43xx driver from loading, so I'm left with either no wifi or no ethenet.

Ok, so I just tested this on a machine and it is _not_ true. It works just fine to run b44/ssb with bcm43xx in parallel.

Quoted 6 lines
Quoted 2 lines

Can you please stop fooling me and spreading this FUD?

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right?

Right. So "so I'm left with either no wifi or no ethenet" being wrong.

Quoted 2 lines

Looks like this, kind of, ruins my statement that is is impossible to get working wifi in 2.6.24. Sorry, it was not intensinal. ;)

Ok, I see.

Quoted 4 lines

But there still are lots of valid combinations where the b43xx driver silently won't work. E.g. when either the ssb or the b44 modules are loaded prior to it, or when either of them are built statically (building the b43xx driver staticaly won't help here). And my patch fixes them.

With your patch you can still artificially create situations that something doesn't work.
My point being: It works just fine without the patch.
And as I'm not really sure about the implications of the patch, I won't ack this. But it seems Larry acked it. So that's also fine, if he takes over responsibility for that change then. So the next "You broke my kernel!!11" round can go to his mail address and I won't be bothered by this anymore. ;)

#22 lore raw

Hi Michael,

Quoted 5 lines · Michael Buesch
Quoted 3 lines

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right?

Right. So "so I'm left with either no wifi or no ethenet" being wrong.

Lets make this simple: it used to work before and now it doesn't. Therefore it's a regression that must be addressed. Period.

Pekka

#23 lore raw
Quoted 11 lines · Pekka J Enberg

Hi Michael,

Quoted 5 lines · Michael Buesch
Quoted 3 lines

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right?

Right. So "so I'm left with either no wifi or no ethenet" being wrong.

Lets make this simple: it used to work before and now it doesn't. Therefore it's a regression that must be addressed. Period.

Isn't the resolution Michael is suggesting is, "use the different driver"?

thanks,

greg k-h

#24 lore raw

Hi Greg,

Quoted 4 lines · Greg KH
Quoted 2 lines

Lets make this simple: it used to work before and now it doesn't. Therefore it's a regression that must be addressed. Period.

Isn't the resolution Michael is suggesting is, "use the different driver"?

Alexey said it didn't work but even if it did, we don't let old drivers _regress_ as long as they're in the tree. There are plenty of precedents here so I don't see why this particular Broadcom driver is any different.

Pekka

#25 lore raw
In reply to Greg KH
Quoted 8 lines · On Monday 25 February 2008 07:49:35 Greg KH wrote:
Quoted 7 lines · Pekka J Enberg

Hi Michael,

Quoted 4 lines · Michael Buesch
Quoted 3 lines

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right?

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

Right. So "so I'm left with either no wifi or no ethenet" being wrong.

Lets make this simple: it used to work before and now it doesn't. Therefore it's a regression that must be addressed. Period.

Isn't the resolution Michael is suggesting is, "use the different driver"?

I have two resolutions. One being:
rmmod b44
rmmod ssb
modprobe bcm43xx
modprobe b44

The other being: Wait for 2.6.25 and use the maintained b43 driver.

Signature · 1 line
Greetings Michael.
#30 lore raw
In reply to Greg KH
Quoted 9 lines · Greg KH
Quoted 7 lines · Pekka J Enberg

Hi Michael,

Quoted 4 lines · Michael Buesch
Quoted 3 lines

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right?

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Right. So "so I'm left with either no wifi or no ethenet" being wrong.

Lets make this simple: it used to work before and now it doesn't. Therefore it's a regression that must be addressed. Period.

Isn't the resolution Michael is suggesting is, "use the different driver"?

The b43 driver from 2.6.24 does not work with my hardware. The one from 2.6.25 seems to work, but 2.6.25 is far from being ready yet.

The only way you can get the old driver working in 2.6.24 is to compile certain (completely unrelated for an outsider) drivers as modules and play with the module loading order. I think this is a bug, and it should be fixed in -stable.

And for 2.6.25, I think the patch should also be included, as the bcm43xx driver is still there.

Btw, what are we discussing? The real bcm43xx maintainer (Larry Finger) reviewed the patch and agreet it should go into 2.6.24-stable and 2.6.25 . Now we only have to wait for John Linville and Jeff Garzik to pick it up, right? If someone still thinks the patch should not go in, he should at least add them to the CC list. ;)

Quoted 4 lines

thanks,

greg k-h

#31 lore raw
Quoted 18 lines · On Monday 25 February 2008 11:23:02 Alexey Zaytsev wrote:
Quoted 9 lines · Greg KH
Quoted 7 lines · Pekka J Enberg

Hi Michael,

Quoted 4 lines · Michael Buesch
Quoted 3 lines

The ony way I see this was possible, you manually changed the module loading order, so that the b43xx module was loaded prior to the ssb and b44 modules. Right?

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Right. So "so I'm left with either no wifi or no ethenet" being wrong.

Lets make this simple: it used to work before and now it doesn't. Therefore it's a regression that must be addressed. Period.

Isn't the resolution Michael is suggesting is, "use the different driver"?

Quoted 7 lines

The b43 driver from 2.6.24 does not work with my hardware. The one from 2.6.25 seems to work, but 2.6.25 is far from being ready yet.

The only way you can get the old driver working in 2.6.24 is to compile certain (completely unrelated for an outsider) drivers as modules and play with the module loading order. I think this is a bug, and it should be fixed in -stable.

It must first go into the 2.6.25 tree and then into -stable. And the patch must be considered to do The Right Thing (tm).

Quoted 9 lines

And for 2.6.25, I think the patch should also be included, as the bcm43xx driver is still there.

Btw, what are we discussing? The real bcm43xx maintainer (Larry Finger) reviewed the patch and agreet it should go into 2.6.24-stable and 2.6.25 . Now we only have to wait for John Linville and Jeff Garzik to pick it up, right? If someone still thinks the patch should not go in, he should at least add them to the CC list. ;)

Heh, wait. Larry is neither the bcm43xx maintainer [1], nor does this code touch bcm43xx code. It does touch ssb and b43 code, which I am the maintainer of.
Though, as I said, if somebody else is familiar with the code does really understand the patch and does ACK it (which Larry did), I'm probably OK with applying it. As long as I'm not bothered with any followup regressions caused by this. So if the one signing off the patch will handle all this, I'm fine with it.

[1] bcm43xx is unmaintained. Larry used to be the maintainer until he dropped it a few months ago.

Signature · 1 line
Greetings Michael.
#32 lore raw
Quoted 2 lines · Michael Buesch

[1] bcm43xx is unmaintained. Larry used to be the maintainer until he dropped it a few months ago.

Doesn't that mean that Alexey gets to be the maintainer, as he's the one patching it ?

Xav

#33 lore raw
Quoted 6 lines · On Monday 25 February 2008 11:49:34 Xavier Bestel wrote:
Quoted 2 lines · Michael Buesch

[1] bcm43xx is unmaintained. Larry used to be the maintainer until he dropped it a few months ago.

Doesn't that mean that Alexey gets to be the maintainer, as he's the one patching it ?

1) He is not patching it.

2) Yeah, if he likes to be the bcm43xx maintainer, please go for it.

Signature · 1 line
Greetings Michael.
#26 lore raw
In reply to Michael Buesch
Quoted 9 lines · Michael Buesch
Quoted 1 line

Isn't the resolution Michael is suggesting is, "use the different driver"?

I have two resolutions. One being:
rmmod b44
rmmod ssb
modprobe bcm43xx
modprobe b44

The other being: Wait for 2.6.25 and use the maintained b43 driver.

Neither of which seem like acceptable solutions for a 2.6.23 -> 2.6.24 _regression_. Or maybe I am just too naive to believe Linus' statement on not letting the kernel regress...

#27 lore raw
Quoted 9 lines · On Monday 25 February 2008 13:11:04 Pekka Enberg wrote:
Quoted 8 lines · Michael Buesch
Quoted 1 line

Isn't the resolution Michael is suggesting is, "use the different driver"?

I have two resolutions. One being:
rmmod b44
rmmod ssb
modprobe bcm43xx
modprobe b44

Quoted 5 lines
Quoted 1 line

The other being: Wait for 2.6.25 and use the maintained b43 driver.

Neither of which seem like acceptable solutions for a 2.6.23 -> 2.6.24 _regression_. Or maybe I am just too naive to believe Linus' statement on not letting the kernel regress...

So, please sign-off the patch that we have, if you think it's right and doesn't cause more regressions.

Signature · 1 line
Greetings Michael.
Reviewed-by
#28 lore raw
Quoted 6 lines · Michael Buesch
Quoted 3 lines

Neither of which seem like acceptable solutions for a 2.6.23 -> 2.6.24 _regression_. Or maybe I am just too naive to believe Linus' statement on not letting the kernel regress...

So, please sign-off the patch that we have, if you think it's right and doesn't cause more regressions.

I did look at the patch and can gladly add a:

But this seems backwards. It was _your_ commit that broke the setup and the patch touches a driver _you're_ maintaining.

So can we just revert commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") from 2.6.24 and you can add it back to 2.6.25 if the problem indeed does go away?

Pekka

#29 lore raw
Quoted 5 lines · Pekka J Enberg
Quoted 4 lines · Michael Buesch
Quoted 3 lines

Neither of which seem like acceptable solutions for a 2.6.23 -> 2.6.24 _regression_. Or maybe I am just too naive to believe Linus' statement on not letting the kernel regress...

Quoted 7 lines
Quoted 2 lines

So, please sign-off the patch that we have, if you think it's right and doesn't cause more regressions.

I did look at the patch and can gladly add a:

Thanks for reviewing. Was it the patch I sent to Larry Finger with the subject line "[PATCH] Fix the bcm43xx driver breakage in 2.6.24/25" or the patch I sent with the first email in this thread? The patches are generally the same, but the one sent to Larry was split into two pieces and the condition on which the bcm43xx config option was hidden changed a bit.

Quoted 6 lines

But this seems backwards. It was _your_ commit that broke the setup and the patch touches a driver _you're_ maintaining.

So can we just revert commit 753f492093da7a40141bfe083073400f518f4c68 ("[B44]: port to native ssb support") from 2.6.24 and you can add it back to 2.6.25 if the problem indeed does go away?

I'm quite sure my patch won't cause any problems, but if Greg wants to be 100% sure there won't be any regressions introduced in -stable, reverting the said commit should be the right thing.

#40 lore raw
In reply to Michael Buesch
Quoted 9 lines · Michael Buesch
Quoted 2 lines · On Saturday 23 February 2008 22:32:46 Alexey Zaytsev wrote:

And you knew that the new driver did no work with the bcm4311 chips, which is the sad thing.

That is not true. It doesn't work with exactly _one_ revision of the bcm4311 card. And that is already fixed in 2.6.25. I'd like to have that in .24-stable, too, but I guess it's too big. It changes some parts of the DMA engine code.

Which patch is it? Is it this one? Or is there another one that touches 4311?

commit 013978b688d2a27af3ab55ca739e8c8ac7254870

Author: Larry Finger <[email protected]>
Date:   Mon Nov 26 10:29:47 2007 -0600

b43: Changes to enable BCM4311 rev 02 with wireless core revision 13

The BCM94311MCG rev 02 chip has an 802.11 core with revision 13 and has not been supported until now. The changes include the following:

<snip>

If so, perhaps we should just consider applying this to -stable?

I really do wish everyone would tone down their rhetoric on this issue...

John

Signature · 2 lines
John W. Linville
[email protected]
#41 lore raw
Quoted 18 lines · John W. Linville
Quoted 10 lines · Michael Buesch
Quoted 2 lines · On Saturday 23 February 2008 22:32:46 Alexey Zaytsev wrote:

And you knew that the new driver did no work with the bcm4311 chips, which is the sad thing.

That is not true. It doesn't work with exactly _one_ revision of the bcm4311 card. And that is already fixed in 2.6.25. I'd like to have that in .24-stable, too, but I guess it's too big. It changes some parts of the DMA engine code.

Which patch is it? Is it this one? Or is there another one that touches 4311?

commit 013978b688d2a27af3ab55ca739e8c8ac7254870
Author: Larry Finger <[email protected]>
Date: Mon Nov 26 10:29:47 2007 -0600

Quoted 9 lines

b43: Changes to enable BCM4311 rev 02 with wireless core revision 13

The BCM94311MCG rev 02 chip has an 802.11 core with revision 13 and has not been supported until now. The changes include the following:

<snip>

If so, perhaps we should just consider applying this to -stable?

I see a few problems here:

This commit seems to depend on some commits prior to it, as it touches drivers/net/wireless/b43/wa.c which is absent in 2.6.24.3. When I get home, I'll test if it still works if you ignore wa.c

The patch may be too complex for -stable. I think there was a hard line count limit, like no more than 100 lines. I'm not sure if this still stands.

This does not fix the bcm43xx driver.

Quoted 8 lines

I really do wish everyone would tone down their rhetoric on this issue...

John
--
John W. Linville
[email protected]

#42 lore raw
Quoted 7 lines · On Tuesday 26 February 2008 22:38:56 Alexey Zaytsev wrote:
Quoted 5 lines · John W. Linville
Quoted 4 lines · Michael Buesch
Quoted 1 line · On Saturday 23 February 2008 22:32:46 Alexey Zaytsev wrote:

And you knew that the new driver did no work with the bcm4311

Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
Quoted 1 line

chips, which is the sad thing.

That is not true. It doesn't work with exactly _one_ revision of the bcm4311 card. And that is already fixed in 2.6.25. I'd like to have that in .24-stable, too, but I guess it's too big. It changes some parts of the DMA engine code.

Which patch is it? Is it this one? Or is there another one that touches 4311?

Yeah, I think that's it.

Quoted 11 lines
Quoted 11 lines

commit 013978b688d2a27af3ab55ca739e8c8ac7254870
Author: Larry Finger <[email protected]>
Date: Mon Nov 26 10:29:47 2007 -0600

b43: Changes to enable BCM4311 rev 02 with wireless core revision 13

The BCM94311MCG rev 02 chip has an 802.11 core with revision 13 and has not been supported until now. The changes include the following:

<snip>

Quoted 8 lines
Quoted 2 lines

If so, perhaps we should just consider applying this to -stable?

I see a few problems here:

This commit seems to depend on some commits prior to it, as it touches drivers/net/wireless/b43/wa.c which is absent in 2.6.24.3. When I get home, I'll test if it still works if you ignore wa.c

Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
First need to find the place in the old phy.c code that matches the new wa.c code.

Quoted 4 lines

The patch may be too complex for -stable. I think there was a hard line count limit, like no more than 100 lines. I'm not sure if this still stands.

This does not fix the bcm43xx driver.

It does not even touch the bcm43xx driver.
Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Signature · 1 line
Greetings Michael.
#43 lore raw
Quoted 40 lines · Michael Buesch
Quoted 7 lines · On Tuesday 26 February 2008 22:38:56 Alexey Zaytsev wrote:
Quoted 5 lines · John W. Linville
Quoted 4 lines · Michael Buesch
Quoted 1 line · On Saturday 23 February 2008 22:32:46 Alexey Zaytsev wrote:

And you knew that the new driver did no work with the bcm4311

Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
Quoted 1 line

chips, which is the sad thing.

That is not true. It doesn't work with exactly _one_ revision of the bcm4311 card. And that is already fixed in 2.6.25. I'd like to have that in .24-stable, too, but I guess it's too big. It changes some parts of the DMA engine code.

Which patch is it? Is it this one? Or is there another one that touches 4311?

Yeah, I think that's it.

Quoted 11 lines
Quoted 11 lines

commit 013978b688d2a27af3ab55ca739e8c8ac7254870
Author: Larry Finger <[email protected]>
Date: Mon Nov 26 10:29:47 2007 -0600

b43: Changes to enable BCM4311 rev 02 with wireless core revision 13

The BCM94311MCG rev 02 chip has an 802.11 core with revision 13 and has not been supported until now. The changes include the following:

<snip>

Quoted 8 lines
Quoted 2 lines

If so, perhaps we should just consider applying this to -stable?

I see a few problems here:

This commit seems to depend on some commits prior to it, as it touches drivers/net/wireless/b43/wa.c which is absent in 2.6.24.3. When I get home, I'll test if it still works if you ignore wa.c

Quoted 6 lines

Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
First need to find the place in the old phy.c code that matches the new wa.c code.

I'll do the testing.

Quoted 10 lines
Quoted 5 lines

The patch may be too complex for -stable. I think there was a hard line count limit, like no more than 100 lines. I'm not sure if this still stands.

This does not fix the bcm43xx driver.

It does not even touch the bcm43xx driver.
Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

Quoted 3 lines

--
Greetings Michael.

#44 lore raw
Quoted 7 lines · On Tuesday 26 February 2008 23:12:32 Alexey Zaytsev wrote:
Quoted 6 lines

 Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
 This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
 First need to find the place in the old phy.c code that matches the new wa.c code.

I'll do the testing.

drivers/net/wireless/b43/dma.c +22 -8
Index: linux-2.6.24.3/drivers/net/wireless/b43/dma.c
===================================================================
@@ -165,7 +165,7 @@ static void op64_fill_descriptor(struct 
 	addrhi = (((u64) dmaaddr >> 32) & ~SSB_DMA_TRANSLATION_MASK);
 	addrext = (((u64) dmaaddr >> 32) & SSB_DMA_TRANSLATION_MASK)
 	    >> SSB_DMA_TRANSLATION_SHIFT;
-	addrhi |= ssb_dma_translation(ring->dev->dev);
+	addrhi |= (ssb_dma_translation(ring->dev->dev) << 1);
 	if (slot == ring->nr_slots - 1)
 		ctl0 |= B43_DMA64_DCTL0_DTABLEEND;
 	if (start)
@@ -426,9 +426,21 @@ static inline
 static int alloc_ringmemory(struct b43_dmaring *ring)
 {
 	struct device *dev = ring->dev->dev->dev;
+	gfp_t flags = GFP_KERNEL;
 
+	/* The specs call for 4K buffers for 30- and 32-bit DMA with 4K
+	 * alignment and 8K buffers for 64-bit DMA with 8K alignment. Testing
+	 * has shown that 4K is sufficient for the latter as long as the buffer
+	 * does not cross an 8K boundary.
+	 *
+	 * For unknown reasons - possibly a hardware error - the BCM4311 rev
+	 * 02, which uses 64-bit DMA, needs the ring buffer in very low memory,
+	 * which accounts for the GFP_DMA flag below.
+	 */
+	if (ring->dma64)
+		flags |= GFP_DMA;
 	ring->descbase = dma_alloc_coherent(dev, B43_DMA_RINGMEMSIZE,
-					    &(ring->dmabase), GFP_KERNEL);
+					    &(ring->dmabase), flags);
 	if (!ring->descbase) {
 		b43err(ring->dev->wl, "DMA ringmemory allocation failed\n");
 		return -ENOMEM;
@@ -483,7 +495,7 @@ int b43_dmacontroller_rx_reset(struct b4
 	return 0;
 }
 
-/* Reset the RX DMA channel */
+/* Reset the TX DMA channel */
 int b43_dmacontroller_tx_reset(struct b43_wldev *dev, u16 mmio_base, int dma64)
 {
 	int i;
@@ -647,7 +659,7 @@ static int dmacontroller_setup(struct b4
 			b43_dma_write(ring, B43_DMA64_TXRINGHI,
 				      ((ringbase >> 32) &
 				       ~SSB_DMA_TRANSLATION_MASK)
-				      | trans);
+				      | (trans << 1));
 		} else {
 			u32 ringbase = (u32) (ring->dmabase);
 
@@ -680,8 +692,9 @@ static int dmacontroller_setup(struct b4
 			b43_dma_write(ring, B43_DMA64_RXRINGHI,
 				      ((ringbase >> 32) &
 				       ~SSB_DMA_TRANSLATION_MASK)
-				      | trans);
-			b43_dma_write(ring, B43_DMA64_RXINDEX, 200);
+				      | (trans << 1));
+			b43_dma_write(ring, B43_DMA64_RXINDEX, ring->nr_slots *
+				      sizeof(struct b43_dmadesc64));
 		} else {
 			u32 ringbase = (u32) (ring->dmabase);
 
@@ -695,11 +708,12 @@ static int dmacontroller_setup(struct b4
 			b43_dma_write(ring, B43_DMA32_RXRING,
 				      (ringbase & ~SSB_DMA_TRANSLATION_MASK)
 				      | trans);
-			b43_dma_write(ring, B43_DMA32_RXINDEX, 200);
+			b43_dma_write(ring, B43_DMA32_RXINDEX, ring->nr_slots *
+				      sizeof(struct b43_dmadesc32));
 		}
 	}
 
-      out:
+out:
 	return err;
 }
 
drivers/net/wireless/b43/main.c +2 -1
Index: linux-2.6.24.3/drivers/net/wireless/b43/main.c
===================================================================
@@ -101,6 +101,7 @@ static const struct ssb_device_id b43_ss
 	SSB_DEVICE(SSB_VENDOR_BROADCOM, SSB_DEV_80211, 7),
 	SSB_DEVICE(SSB_VENDOR_BROADCOM, SSB_DEV_80211, 9),
 	SSB_DEVICE(SSB_VENDOR_BROADCOM, SSB_DEV_80211, 10),
+	SSB_DEVICE(SSB_VENDOR_BROADCOM, SSB_DEV_80211, 13),
 	SSB_DEVTABLE_END
 };
 
@@ -3079,7 +3080,7 @@ static int b43_phy_versioning(struct b43
 			unsupported = 1;
 		break;
 	case B43_PHYTYPE_G:
-		if (phy_rev > 8)
+		if (phy_rev > 9)
 			unsupported = 1;
 		break;
 	default:
Signature · 1 line
Greetings Michael.
#50 lore raw
In reply to Alexey Zaytsev
Quoted 7 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

John

Signature · 2 lines
John W. Linville
[email protected]
#51 lore raw
Quoted 20 lines · On Tuesday 26 February 2008 23:47:57 John W. Linville wrote:
Quoted 7 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Quoted 3 lines

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

You got the point exactly right John. :)
The special "problem" about this is that the b44 driver, which _seems_ to have nothing to do with b43, registers the b43 PCI IDs. (This is just for convenience. See the comment in the header of b43_pci_bridge.c for details).
So if you use the b44 driver, but also want to use the bcm43xx driver, you have to load the bcm43xx driver first to make it probe before b44 registers the b43 PCI IDs.

The bcm43xx driver is not broken _at_ _all_. In fact, its code did not even slightly change since ages.

I do see the issue, but I don't feel like changing the complicated SSB KConfig stuff (it has lots of weird dependencies) to fix this bug that is actually trivial to work around. Besides that, how many people do we have, which do have a b44 card plus a bcm4311 rev 1 (there are _LOTS_ of revisions of the 43xx hardware)?

So I'd rather fix b43 to work on the card. See the patch I just sent.

Signature · 1 line
Greetings Michael.
#52 lore raw
In reply to John W. Linville
Quoted 21 lines · John W. Linville
Quoted 8 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Quoted 3 lines

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

Is there any reason the ssb driver should claim the bcm43xx pci ids in the first place? I have very little understanding what the Sonic Silicon Backplane really is, but I see that the b44 driver claims its PCI ids directly. I also think I understand why the b43/b43legacy drivers can't claim the ids directly: because the driver-device matching is done not with the pci bus methods, but with the ssb bus methods, and it would be impossible to automatically choose the right driver for the right device (with same ssb ids), as the first of the two drivers loaded would succeed in probe()'ing the pci "ssb bridge" device, and not letting the other to take control, even after moments later the ssb probe for the non-supported ssb device would fail. (Or am I completely wrong?)

That said, I still think that the ssb driver claims the wrong pci ids, which is especially wrong if the the b43/b43legacy drivers are not even built. And my patch fixes exactly this problem - the ssb driver no more claims the broadcom pci ids, when the b43/b43legacy drivers are not built.

One better solution I think might be to move the b43_pci_bridge.c code to a separate module, and let the b43/b43legacy drivers depend on it, but as I said, I have little knowledge in the ssb stuff, so I did it the easy way.

Btw, at least in the 8139cp/8139too case the right driver is chosen automatically, because they both can detect the problem at the pci probe() time and tell the device is not supported early.

Quoted 8 lines

John
--

John W. Linville
[email protected]

#53 lore raw
Quoted 45 lines · On Wednesday 27 February 2008 01:23:17 Alexey Zaytsev wrote:
Quoted 21 lines · John W. Linville
Quoted 8 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Quoted 3 lines

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

Is there any reason the ssb driver should claim the bcm43xx pci ids in the first place? I have very little understanding what the Sonic Silicon Backplane really is, but I see that the b44 driver claims its PCI ids directly. I also think I understand why the b43/b43legacy drivers can't claim the ids directly: because the driver-device matching is done not with the pci bus methods, but with the ssb bus methods, and it would be impossible to automatically choose the right driver for the right device (with same ssb ids), as the first of the two drivers loaded would succeed in probe()'ing the pci "ssb bridge" device, and not letting the other to take control, even after moments later the ssb probe for the non-supported ssb device would fail. (Or am I completely wrong?)

That said, I still think that the ssb driver claims the wrong pci ids, which is especially wrong if the the b43/b43legacy drivers are not even built. And my patch fixes exactly this problem - the ssb driver no more claims the broadcom pci ids, when the b43/b43legacy drivers are not built.

Quoted 4 lines

One better solution I think might be to move the b43_pci_bridge.c code to a separate module, and let the b43/b43legacy drivers depend on it, but as I said, I have little knowledge in the ssb stuff, so I did it the easy way.

See the comment in b43_pci_bridge.c

Signature · 1 line
Greetings Michael.
#54 lore raw
Quoted 47 lines · Michael Buesch
Quoted 45 lines · On Wednesday 27 February 2008 01:23:17 Alexey Zaytsev wrote:
Quoted 21 lines · John W. Linville
Quoted 8 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Quoted 3 lines

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

Is there any reason the ssb driver should claim the bcm43xx pci ids in the first place? I have very little understanding what the Sonic Silicon Backplane really is, but I see that the b44 driver claims its PCI ids directly. I also think I understand why the b43/b43legacy drivers can't claim the ids directly: because the driver-device matching is done not with the pci bus methods, but with the ssb bus methods, and it would be impossible to automatically choose the right driver for the right device (with same ssb ids), as the first of the two drivers loaded would succeed in probe()'ing the pci "ssb bridge" device, and not letting the other to take control, even after moments later the ssb probe for the non-supported ssb device would fail. (Or am I completely wrong?)

That said, I still think that the ssb driver claims the wrong pci ids, which is especially wrong if the the b43/b43legacy drivers are not even built. And my patch fixes exactly this problem - the ssb driver no more claims the broadcom pci ids, when the b43/b43legacy drivers are not built.

Quoted 7 lines
Quoted 4 lines

One better solution I think might be to move the b43_pci_bridge.c code to a separate module, and let the b43/b43legacy drivers depend on it, but as I said, I have little knowledge in the ssb stuff, so I did it the easy way.

See the comment in b43_pci_bridge.c

Yes, I've seen it. And this design, kind of, causes me some trouble. ^

Quoted 3 lines

--
Greetings Michael.

#55 lore raw
Quoted 48 lines · On Wednesday 27 February 2008 01:32:21 Alexey Zaytsev wrote:
Quoted 47 lines · Michael Buesch
Quoted 45 lines · On Wednesday 27 February 2008 01:23:17 Alexey Zaytsev wrote:
Quoted 21 lines · John W. Linville
Quoted 8 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Quoted 3 lines

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

Is there any reason the ssb driver should claim the bcm43xx pci ids in the first place? I have very little understanding what the Sonic Silicon Backplane really is, but I see that the b44 driver claims its PCI ids directly. I also think I understand why the b43/b43legacy drivers can't claim the ids directly: because the driver-device matching is done not with the pci bus methods, but with the ssb bus methods, and it would be impossible to automatically choose the right driver for the right device (with same ssb ids), as the first of the two drivers loaded would succeed in probe()'ing the pci "ssb bridge" device, and not letting the other to take control, even after moments later the ssb probe for the non-supported ssb device would fail. (Or am I completely wrong?)

That said, I still think that the ssb driver claims the wrong pci ids, which is especially wrong if the the b43/b43legacy drivers are not even built. And my patch fixes exactly this problem - the ssb driver no more claims the broadcom pci ids, when the b43/b43legacy drivers are not built.

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines

One better solution I think might be to move the b43_pci_bridge.c code to a separate module, and let the b43/b43legacy drivers depend on it, but as I said, I have little knowledge in the ssb stuff, so I did it the easy way.

See the comment in b43_pci_bridge.c

Yes, I've seen it. And this design, kind of, causes me some trouble.

There are several reasons to not do a seperate module.

First one being: People won't load it and complain about a regression. Yeah, stupid stuff like that happens all the time. That's the reason we SELECT the ssb code instead of using DEPENDS ON. People will otherwise not enable it and report regressions.
And please don't say this won't happen. It _did_ happen when b44-PCI was a seperate kconfig option. People reported regressions, although there were none.

Second one being: It's overkill to have a seperate module for two lines of code and a PCI ID table.

Third one being: We want the code to be as small as possible, as it must all run on embedded machines. In general being as small as possible should be the way to go.

Signature · 1 line
Greetings Michael.
#56 lore raw
Quoted 59 lines · Michael Buesch
Quoted 48 lines · On Wednesday 27 February 2008 01:32:21 Alexey Zaytsev wrote:
Quoted 47 lines · Michael Buesch
Quoted 45 lines · On Wednesday 27 February 2008 01:23:17 Alexey Zaytsev wrote:
Quoted 21 lines · John W. Linville
Quoted 8 lines · Alexey Zaytsev
Quoted 3 lines · Michael Buesch

Besides that the bcm43xx driver is not broken. That's the whole reason this damn thread started at all. So it can't be broken.

Can't agree here. The bcm43xx driver used to work with 2.6.23 without requiring any module magic.

At the risk of prolonging things... :-(

Isn't the fundamental problem here that the ssb driver claims the same PCI IDs as the bcm43xx driver? He have hit this same issue a number of times: 8139too vs. 8139cp, eepro vs. e100, sk98lin vs. skge, and I'm sure there are more. I admit that this situation is a bit more confusing, since the user is less likely to predict a conflict between bcm43xx and the ssb driver. This is especially true since the user isn't even selecting ssb directly, but is instead selecting the apparently unrelated b44.

Quoted 3 lines

Still, the bcm43xx driver is not fundamentally damaged. This is fundamentally a "two drivers claiming the same PCI ID" issue, not a "you broke my driver" one.

Is there any reason the ssb driver should claim the bcm43xx pci ids in the first place? I have very little understanding what the Sonic Silicon Backplane really is, but I see that the b44 driver claims its PCI ids directly. I also think I understand why the b43/b43legacy drivers can't claim the ids directly: because the driver-device matching is done not with the pci bus methods, but with the ssb bus methods, and it would be impossible to automatically choose the right driver for the right device (with same ssb ids), as the first of the two drivers loaded would succeed in probe()'ing the pci "ssb bridge" device, and not letting the other to take control, even after moments later the ssb probe for the non-supported ssb device would fail. (Or am I completely wrong?)

That said, I still think that the ssb driver claims the wrong pci ids, which is especially wrong if the the b43/b43legacy drivers are not even built. And my patch fixes exactly this problem - the ssb driver no more claims the broadcom pci ids, when the b43/b43legacy drivers are not built.

Quoted 8 lines
Quoted 7 lines
Quoted 4 lines

One better solution I think might be to move the b43_pci_bridge.c code to a separate module, and let the b43/b43legacy drivers depend on it, but as I said, I have little knowledge in the ssb stuff, so I did it the easy way.

See the comment in b43_pci_bridge.c

Yes, I've seen it. And this design, kind of, causes me some trouble.

Quoted 9 lines

There are several reasons to not do a seperate module.

First one being: People won't load it and complain about a regression. Yeah, stupid stuff like that happens all the time. That's the reason we SELECT the ssb code instead of using DEPENDS ON. People will otherwise not enable it and report regressions.
And please don't say this won't happen. It _did_ happen when b44-PCI was a seperate kconfig option. People reported regressions, although there were none.

I mean, not to make it a separate config option, of course it should be hidden and
autoselected, but a separate module with maybe a symbol both b43/b43legacy could depend on.

Quoted 7 lines

Second one being: It's overkill to have a seperate module for two lines of code and a PCI ID table.

Third one being: We want the code to be as small as possible, as it must all run on embedded machines. In general being as small as possible should be the way to go.

If compiled statically, which I think is the right thing for embedded systems, it should not take much additional space when separated from the ssb code.

#45 lore raw
In reply to Michael Buesch
Quoted 14 lines · On Tuesday 26 February 2008 23:41:55 Michael Buesch wrote:
Quoted 7 lines · On Tuesday 26 February 2008 23:12:32 Alexey Zaytsev wrote:
Quoted 6 lines

 Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
 This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
 First need to find the place in the old phy.c code that matches the new wa.c code.

I'll do the testing.

Index: linux-2.6.24.3/drivers/net/wireless/b43/dma.c
===================================================================

drivers/net/wireless/b43/dma.c static void op64_fill_descriptor(struct · line 166

Quoted 2 lines
drivers/net/wireless/b43/dma.c +0 -0
@@ -165,7 +165,7 @@ static void op64_fill_descriptor(struct 
 	addrhi = (((u64) dmaaddr >> 32) & ~SSB_DMA_TRANSLATION_MASK);
 	addrext = (((u64) dmaaddr >> 32) & SSB_DMA_TRANSLATION_MASK)

[snip]

So are you actually going to do this anytime soon?

Signature · 1 line
Greetings Michael.
#46 lore raw
Quoted 19 lines · Michael Buesch
Quoted 16 lines · On Tuesday 26 February 2008 23:41:55 Michael Buesch wrote:
Quoted 7 lines · On Tuesday 26 February 2008 23:12:32 Alexey Zaytsev wrote:
Quoted 6 lines

Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
First need to find the place in the old phy.c code that matches the new wa.c code.

I'll do the testing.

Index: linux-2.6.24.3/drivers/net/wireless/b43/dma.c
===================================================================

drivers/net/wireless/b43/dma.c 2008-02-26 23:39:43.000000000 +0100 renamed from drivers/net/wireless/b43/dma.c 2008-02-26 23:39:39.000000000 +0100 +0 -0
@@ -165,7 +165,7 @@ static void op64_fill_descriptor(struct
      addrhi = (((u64) dmaaddr >> 32) & ~SSB_DMA_TRANSLATION_MASK);
      addrext = (((u64) dmaaddr >> 32) & SSB_DMA_TRANSLATION_MASK)

[snip]

Quoted 2 lines

So are you actually going to do this anytime soon?

Sorry for the delay. Yes, it worked fine.

Btw, when booting the system with the b43 driver, I get a ~30 socond delay after the boot scripts report "Waiting for /dev to be fully populated" and all the modules get loaded. Is this something expected?

Quoted 3 lines

--
Greetings Michael.

#47 lore raw

Hi

Quoted 11 lines · Alexey Zaytsev
Quoted 10 lines · Michael Buesch
Quoted 9 lines · On Tuesday 26 February 2008 23:41:55 Michael Buesch wrote:
Quoted 7 lines · On Tuesday 26 February 2008 23:12:32 Alexey Zaytsev wrote:
Quoted 6 lines

Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
First need to find the place in the old phy.c code that matches the new wa.c code.

I'll do the testing.

[...]

Quoted 10 lines
Quoted 3 lines

So are you actually going to do this anytime soon?

Sorry for the delay. Yes, it worked fine.

Btw, when booting the system with the b43 driver, I get a ~30 socond delay after the boot scripts report "Waiting for /dev to be fully populated" and all the modules get loaded. Is this something expected?

That is most likely udev related, which doesn't really like module names and default interface names changing for a known MAC address, depending on your distribution (the following example assumes debian sid, udev 0.114-2) you might have something like /etc/udev/rules.d/z25_persistent-net.rules (which lists MAC addresses and the associated persistent interface name), removing the corresponding stanza for your Broadcom wireless card (or removing the whole file alltogether and letting udev (or rather z45_persistent-net-generator.rules) recreate it during the next boot will most likely fix it.

This issue is in no way b43 related, it happens with "every" networking devices with changing module names (at least for every legacy wireless module being replaced by a new mac80211 based driver).

Regards
Stefan Lippers-Hollmann

#49 lore raw
In reply to Alexey Zaytsev
Quoted 20 lines · On Thursday 28 February 2008 23:48:48 Alexey Zaytsev wrote:
Quoted 19 lines · Michael Buesch
Quoted 16 lines · On Tuesday 26 February 2008 23:41:55 Michael Buesch wrote:
Quoted 7 lines · On Tuesday 26 February 2008 23:12:32 Alexey Zaytsev wrote:
Quoted 6 lines

Yeah, seems so. But if you are willing to test an adjusted version, I can probably do a backported version of this patch.
This patch is well tested and does the right thing, so I'm willing to sign off a backported version of this.
First need to find the place in the old phy.c code that matches the new wa.c code.

I'll do the testing.

Index: linux-2.6.24.3/drivers/net/wireless/b43/dma.c
===================================================================

drivers/net/wireless/b43/dma.c 2008-02-26 23:39:43.000000000 +0100 renamed from drivers/net/wireless/b43/dma.c 2008-02-26 23:39:39.000000000 +0100 +0 -0
@@ -165,7 +165,7 @@ static void op64_fill_descriptor(struct
      addrhi = (((u64) dmaaddr >> 32) & ~SSB_DMA_TRANSLATION_MASK);
      addrext = (((u64) dmaaddr >> 32) & SSB_DMA_TRANSLATION_MASK)

[snip]

Quoted 4 lines
Quoted 2 lines

So are you actually going to do this anytime soon?

Sorry for the delay. Yes, it worked fine.

Fine :)

Quoted 4 lines

Btw, when booting the system with the b43 driver, I get a ~30 socond delay after the boot scripts report "Waiting for /dev to be fully populated" and all the modules get loaded. Is this something expected?

No. Probably firmware related.

Signature · 1 line
Greetings Michael.
#48 lore raw
In reply to Stefan Lippers-Hollmann

On Fri, Feb 29, 2008 at 2:12 AM, Stefan Lippers-Hollmann <[email protected]> wrote:
[...]

Quoted 16 lines
Quoted 5 lines

Btw, when booting the system with the b43 driver, I get a ~30 socond delay after the boot scripts report "Waiting for /dev to be fully populated" and all the modules get loaded. Is this something expected?

That is most likely udev related, which doesn't really like module names and default interface names changing for a known MAC address, depending on your distribution (the following example assumes debian sid, udev 0.114-2) you might have something like /etc/udev/rules.d/z25_persistent-net.rules (which lists MAC addresses and the associated persistent interface name), removing the corresponding stanza for your Broadcom wireless card (or removing the whole file alltogether and letting udev (or rather z45_persistent-net-generator.rules) recreate it during the next boot will most likely fix it.

Quoted 3 lines

This issue is in no way b43 related, it happens with "every" networking devices with changing module names (at least for every legacy wireless module being replaced by a new mac80211 based driver).

Removing the file helped. Thanks.

Quoted 4 lines

Regards
Stefan Lippers-Hollmann