iwlwifi: work around reverse dependency on MEI

Message ID	20211207125430.2423871-1-arnd@kernel.org (mailing list archive)
State	Superseded
Delegated to:	Netdev Maintainers
Headers	show Return-Path: <netdev-owner@kernel.org> From: Arnd Bergmann <arnd@kernel.org> To: Luca Coelho <luciano.coelho@intel.com>, Kalle Valo <kvalo@codeaurora.org> Cc: Arnd Bergmann <arnd@arndb.de>, "David S. Miller" <davem@davemloft.net>, Jakub Kicinski <kuba@kernel.org>, Emmanuel Grumbach <emmanuel.grumbach@intel.com>, Ayala Beker <ayala.beker@intel.com>, Miri Korenblit <miriam.rachel.korenblit@intel.com>, Johannes Berg <johannes.berg@intel.com>, linux-wireless@vger.kernel.org, netdev@vger.kernel.org, linux-kernel@vger.kernel.org Subject: [PATCH] iwlwifi: work around reverse dependency on MEI Date: Tue, 7 Dec 2021 13:54:12 +0100 Message-Id: <20211207125430.2423871-1-arnd@kernel.org> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: bulk
Series	iwlwifi: work around reverse dependency on MEI \| expand iwlwifi: work around reverse dependency on MEI

Context	Check	Description
netdev/tree_selection	success	Not a local patch

Arnd Bergmann Dec. 7, 2021, 12:54 p.m. UTC

From: Arnd Bergmann <arnd@arndb.de>

If the iwlmei code is a loadable module, the main iwlwifi driver
cannot be built-in:

x86_64-linux-ld: drivers/net/wireless/intel/iwlwifi/pcie/trans.o: in function `iwl_pcie_prepare_card_hw':
trans.c:(.text+0x4158): undefined reference to `iwl_mei_is_connected'

Unfortunately, Kconfig enforces the opposite, forcing the MEI driver to
not be built-in if iwlwifi is a module.

There is no easy way to express the correct dependency in Kconfig,
this is the best workaround I could come up with, turning CONFIG_IWLMEI
into a 'bool' symbol, and spelling out the exact conditions under which
it may be enabled, and then using Makefile logic to ensure it is
built-in when iwlwifi is.

A better option would be change iwl_mei_is_connected() so it could be
called from iwlwifi regardless of whether the mei driver is reachable,
but that requires a larger rework in the driver.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/net/wireless/intel/iwlwifi/Kconfig      | 6 +++---
 drivers/net/wireless/intel/iwlwifi/Makefile     | 3 +--
 drivers/net/wireless/intel/iwlwifi/mei/Makefile | 4 +++-
 3 files changed, 7 insertions(+), 6 deletions(-)

Emmanuel Grumbach Dec. 7, 2021, 1:19 p.m. UTC | #1

Hi Arend,

> Subject: [PATCH] iwlwifi: work around reverse dependency on MEI
> 
> From: Arnd Bergmann <arnd@arndb.de>
> 
> If the iwlmei code is a loadable module, the main iwlwifi driver cannot be
> built-in:
> 
> x86_64-linux-ld: drivers/net/wireless/intel/iwlwifi/pcie/trans.o: in function
> `iwl_pcie_prepare_card_hw':
> trans.c:(.text+0x4158): undefined reference to `iwl_mei_is_connected'
> 
> Unfortunately, Kconfig enforces the opposite, forcing the MEI driver to not
> be built-in if iwlwifi is a module.
> 
> There is no easy way to express the correct dependency in Kconfig, this is the
> best workaround I could come up with, turning CONFIG_IWLMEI into a 'bool'
> symbol, and spelling out the exact conditions under which it may be enabled,
> and then using Makefile logic to ensure it is built-in when iwlwifi is.
> 
> A better option would be change iwl_mei_is_connected() so it could be
> called from iwlwifi regardless of whether the mei driver is reachable, but that
> requires a larger rework in the driver.

I can try to do that but I don't really see how..
I can't really make a function that would behave differently based on whether the symbol is available or not.

> 
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  drivers/net/wireless/intel/iwlwifi/Kconfig      | 6 +++---
>  drivers/net/wireless/intel/iwlwifi/Makefile     | 3 +--
>  drivers/net/wireless/intel/iwlwifi/mei/Makefile | 4 +++-
>  3 files changed, 7 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/wireless/intel/iwlwifi/Kconfig
> b/drivers/net/wireless/intel/iwlwifi/Kconfig
> index cf1125d84929..474afc6f82a8 100644
> --- a/drivers/net/wireless/intel/iwlwifi/Kconfig
> +++ b/drivers/net/wireless/intel/iwlwifi/Kconfig
> @@ -93,10 +93,10 @@ config IWLWIFI_BCAST_FILTERING
>  	  expect incoming broadcasts for their normal operations.
> 
>  config IWLMEI
> -	tristate "Intel Management Engine communication over WLAN"
> -	depends on INTEL_MEI
> +	bool "Intel Management Engine communication over WLAN"
> +	depends on INTEL_MEI=y || INTEL_MEI=IWLMVM
> +	depends on IWLMVM=y || IWLWIFI=m
>  	depends on PM
> -	depends on IWLMVM
>  	help
>  	  Enables the iwlmei kernel module.

Johannes suggested to make IWLMVM depend on IWLMEI || !IWLMEI
That worked as well, I just had issues with this in our internal backport based tree.
I need to spend a bit more time on this, but I admit my total ignorance in Kconfig's dialect.


> 
> diff --git a/drivers/net/wireless/intel/iwlwifi/Makefile
> b/drivers/net/wireless/intel/iwlwifi/Makefile
> index 75a703eb1bdf..c117e105fe5c 100644
> --- a/drivers/net/wireless/intel/iwlwifi/Makefile
> +++ b/drivers/net/wireless/intel/iwlwifi/Makefile
> @@ -29,7 +29,6 @@ iwlwifi-$(CONFIG_IWLWIFI_DEVICE_TRACING) += iwl-
> devtrace.o  ccflags-y += -I$(src)
> 
>  obj-$(CONFIG_IWLDVM)	+= dvm/
> -obj-$(CONFIG_IWLMVM)	+= mvm/
> -obj-$(CONFIG_IWLMEI)	+= mei/
> +obj-$(CONFIG_IWLMVM)	+= mvm/ mei/
> 
>  CFLAGS_iwl-devtrace.o := -I$(src)
> diff --git a/drivers/net/wireless/intel/iwlwifi/mei/Makefile
> b/drivers/net/wireless/intel/iwlwifi/mei/Makefile
> index 8e3ef0347db7..98b561c3820f 100644
> --- a/drivers/net/wireless/intel/iwlwifi/mei/Makefile
> +++ b/drivers/net/wireless/intel/iwlwifi/mei/Makefile
> @@ -1,5 +1,7 @@
>  # SPDX-License-Identifier: GPL-2.0-only
> -obj-$(CONFIG_IWLMEI)	+= iwlmei.o
> +ifdef CONFIG_IWLMEI
> +obj-$(CONFIG_IWLWIFI)	+= iwlmei.o
> +endif
>  iwlmei-y += main.o
>  iwlmei-y += net.o
>  iwlmei-$(CONFIG_IWLWIFI_DEVICE_TRACING) += trace.o
> --
> 2.29.2

Arnd Bergmann Dec. 7, 2021, 1:25 p.m. UTC | #2

On Tue, Dec 7, 2021 at 2:19 PM Grumbach, Emmanuel
<emmanuel.grumbach@intel.com> wrote:
> > A better option would be change iwl_mei_is_connected() so it could be
> > called from iwlwifi regardless of whether the mei driver is reachable, but that
> > requires a larger rework in the driver.
>
> I can try to do that but I don't really see how..
> I can't really make a function that would behave differently based on whether the symbol is available or not.

I meant that this would be an inline function that only accesses variables
that are available to the iwlwifi driver already, rather than part of the iwlmei
driver.

Part of the problem here is that the current implementation checks a global
variable that is part of the iwlmei driver, so there is no easy way for iwlwifi
to access it.

> >  config IWLMEI
> > -     tristate "Intel Management Engine communication over WLAN"
> > -     depends on INTEL_MEI
> > +     bool "Intel Management Engine communication over WLAN"
> > +     depends on INTEL_MEI=y || INTEL_MEI=IWLMVM
> > +     depends on IWLMVM=y || IWLWIFI=m
> >       depends on PM
> > -     depends on IWLMVM
> >       help
> >         Enables the iwlmei kernel module.
>
> Johannes suggested to make IWLMVM depend on IWLMEI || !IWLMEI
> That worked as well, I just had issues with this in our internal backport based tree.
> I need to spend a bit more time on this, but I admit my total ignorance in Kconfig's dialect.

It's still not enough, the dependency is in iwlwifi, not in iwlmvm, so it
would remain broken for IWLWIFI=y IWLMVM=m IWLMEI=m.

One easy solution might be to link all the iwlmei objects intro the main
iwlwifi driver, the same way it links in the fw/*.c files.

       Arnd

Arnd Bergmann Dec. 7, 2021, 2:10 p.m. UTC | #3

On Tue, Dec 7, 2021 at 1:54 PM Arnd Bergmann <arnd@kernel.org> wrote:
> diff --git a/drivers/net/wireless/intel/iwlwifi/Kconfig b/drivers/net/wireless/intel/iwlwifi/Kconfig
> index cf1125d84929..474afc6f82a8 100644
> --- a/drivers/net/wireless/intel/iwlwifi/Kconfig
> +++ b/drivers/net/wireless/intel/iwlwifi/Kconfig
> @@ -93,10 +93,10 @@ config IWLWIFI_BCAST_FILTERING
>           expect incoming broadcasts for their normal operations.
>
>  config IWLMEI
> -       tristate "Intel Management Engine communication over WLAN"
> -       depends on INTEL_MEI
> +       bool "Intel Management Engine communication over WLAN"
> +       depends on INTEL_MEI=y || INTEL_MEI=IWLMVM
> +       depends on IWLMVM=y || IWLWIFI=m

For reference, that line is wrong, and I still see the same problem
with my patch
applied. It should work after changing it to

        depends on IWLMVM=y || IWLMVM=m

but I'm still testing for further problems.

             Arnd

Johannes Berg Dec. 7, 2021, 2:40 p.m. UTC | #4

On Tue, 2021-12-07 at 14:25 +0100, Arnd Bergmann wrote:

> > >  config IWLMEI
> > > -     tristate "Intel Management Engine communication over WLAN"
> > > -     depends on INTEL_MEI
> > > +     bool "Intel Management Engine communication over WLAN"
> > > +     depends on INTEL_MEI=y || INTEL_MEI=IWLMVM
> > > +     depends on IWLMVM=y || IWLWIFI=m
> > >       depends on PM
> > > -     depends on IWLMVM
> > >       help
> > >         Enables the iwlmei kernel module.
> > 
> > Johannes suggested to make IWLMVM depend on IWLMEI || !IWLMEI
> > That worked as well, I just had issues with this in our internal backport based tree.
> > I need to spend a bit more time on this, but I admit my total ignorance in Kconfig's dialect.
> 
> It's still not enough, the dependency is in iwlwifi, not in iwlmvm, so it
> would remain broken for IWLWIFI=y IWLMVM=m IWLMEI=m.
> 

I missed the pcie/trans.c dependency, and the others are (I think) in
mvm...

but then we can do

config IWLWIFI
	...
	depends on IWLMEI || !IWLMEI
	...

no? That way, we exclude IWLWIFI=y && IWLMEI=m, which I believe causes
the issue? And IWLMVM already depends on IWLWIFI (via the if clause), so
that 

johannes

Arnd Bergmann Dec. 7, 2021, 2:56 p.m. UTC | #5

On Tue, Dec 7, 2021 at 3:40 PM Johannes Berg <johannes@sipsolutions.net> wrote:
>
> On Tue, 2021-12-07 at 14:25 +0100, Arnd Bergmann wrote:
>
> > > >  config IWLMEI
> > > > -     tristate "Intel Management Engine communication over WLAN"
> > > > -     depends on INTEL_MEI
> > > > +     bool "Intel Management Engine communication over WLAN"
> > > > +     depends on INTEL_MEI=y || INTEL_MEI=IWLMVM
> > > > +     depends on IWLMVM=y || IWLWIFI=m
> > > >       depends on PM
> > > > -     depends on IWLMVM
> > > >       help
> > > >         Enables the iwlmei kernel module.
> > >
> > > Johannes suggested to make IWLMVM depend on IWLMEI || !IWLMEI
> > > That worked as well, I just had issues with this in our internal backport based tree.
> > > I need to spend a bit more time on this, but I admit my total ignorance in Kconfig's dialect.
> >
> > It's still not enough, the dependency is in iwlwifi, not in iwlmvm, so it
> > would remain broken for IWLWIFI=y IWLMVM=m IWLMEI=m.
> >
>
> I missed the pcie/trans.c dependency, and the others are (I think) in
> mvm...
>
> but then we can do
>
> config IWLWIFI
>         ...
>         depends on IWLMEI || !IWLMEI
>         ...
>
> no? That way, we exclude IWLWIFI=y && IWLMEI=m, which I believe causes
> the issue? And IWLMVM already depends on IWLWIFI (via the if clause), so
> that

Right, that should work. Testing with that version now.

         Arnd

iwlwifi: work around reverse dependency on MEI

Checks

Commit Message

Comments

Patch