[PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Keith Busch]

From: Keith Busch <kbusch@kernel.org>

Commit e7e222ad73d9 moves devm_cxl_add_nvdimm_bridge() into the cxl_pmem
file, which has independent config compile options for built-in or
module. The call from cxl_acpi_probe() is guarded by
IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.

When CONFIG_CXL_PMEM=m, a built-in cxl_acpi attempts to reference a
symbol exported by a module, which fails to link. CXL_PMEM cannot simply
be promoted to =y in this configuration because it depends on LIBNVDIMM,
which may itself be =m.

Add a Kconfig dependency to prevent CXL_ACPI from being built-in when
CXL_PMEM is a module. This contrains CXL_ACPI to =m when CXL_PMEM=m,
while still allowing CXL_ACPI to be freely configured when CXL_PMEM is
either built-in or disabled.

Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
Signed-off-by: Keith Busch <kbusch@kernel.org>
---
 drivers/cxl/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
index 4589bf11d3fe0..80aeb0d556bd7 100644
--- a/drivers/cxl/Kconfig
+++ b/drivers/cxl/Kconfig
@@ -59,6 +59,7 @@ config CXL_ACPI
 	tristate "CXL ACPI: Platform Support"
 	depends on ACPI
 	depends on ACPI_NUMA
+	depends on CXL_PMEM || !CXL_PMEM
 	default CXL_BUS
 	select ACPI_TABLE_LIB
 	select ACPI_HMAT
-- 
2.47.3

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Gregory Price]

On Thu, Mar 05, 2026 at 12:40:56PM -0800, Keith Busch wrote:
> From: Keith Busch <kbusch@kernel.org>
> 
> Commit e7e222ad73d9 moves devm_cxl_add_nvdimm_bridge() into the cxl_pmem
> file, which has independent config compile options for built-in or
> module. The call from cxl_acpi_probe() is guarded by
> IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.
> 
> When CONFIG_CXL_PMEM=m, a built-in cxl_acpi attempts to reference a
> symbol exported by a module, which fails to link. CXL_PMEM cannot simply
> be promoted to =y in this configuration because it depends on LIBNVDIMM,
> which may itself be =m.
> 
> Add a Kconfig dependency to prevent CXL_ACPI from being built-in when
> CXL_PMEM is a module. This contrains CXL_ACPI to =m when CXL_PMEM=m,
> while still allowing CXL_ACPI to be freely configured when CXL_PMEM is
> either built-in or disabled.
> 
> Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> Signed-off-by: Keith Busch <kbusch@kernel.org>
> ---
>  drivers/cxl/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
> index 4589bf11d3fe0..80aeb0d556bd7 100644
> --- a/drivers/cxl/Kconfig
> +++ b/drivers/cxl/Kconfig
> @@ -59,6 +59,7 @@ config CXL_ACPI
>  	tristate "CXL ACPI: Platform Support"
>  	depends on ACPI
>  	depends on ACPI_NUMA
> +	depends on CXL_PMEM || !CXL_PMEM

Trivially looking at this it looks funky lol

alternatively:

https://lore.kernel.org/linux-cxl/20260302200429.803417-1-gourry@gourry.net/

Not sure which one we want here

~Gregory

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Keith Busch]

On Thu, Mar 05, 2026 at 05:12:31PM -0500, Gregory Price wrote:
> > @@ -59,6 +59,7 @@ config CXL_ACPI
> >  	tristate "CXL ACPI: Platform Support"
> >  	depends on ACPI
> >  	depends on ACPI_NUMA
> > +	depends on CXL_PMEM || !CXL_PMEM
> 
> Trivially looking at this it looks funky lol
 
> alternatively:
> 
> https://lore.kernel.org/linux-cxl/20260302200429.803417-1-gourry@gourry.net/
> 
> Not sure which one we want here

Yeah, I found your proposal just minutes after sending mine.

I agree this is whacky syntax, but it's a somewhat common Kconfig
pattern (see NF_CONNTRACK in net/ipv4/netfilter/Kconfig).

Anyway, my proposal here should force demotion to "m" for CXL_ACPI if
PMEM is also "m", where your proposal should force a failure path at
compile time if ACPI=y and PMEM=m. I really don't know if there's any
preference at all, I also just wanted the kernel build to complete
successfuly. :)

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Gregory Price]

On Thu, Mar 05, 2026 at 03:43:17PM -0700, Keith Busch wrote:
> On Thu, Mar 05, 2026 at 05:12:31PM -0500, Gregory Price wrote:
> > > @@ -59,6 +59,7 @@ config CXL_ACPI
> > >  	tristate "CXL ACPI: Platform Support"
> > >  	depends on ACPI
> > >  	depends on ACPI_NUMA
> > > +	depends on CXL_PMEM || !CXL_PMEM
> > 
> > Trivially looking at this it looks funky lol
>  
> > alternatively:
> > 
> > https://lore.kernel.org/linux-cxl/20260302200429.803417-1-gourry@gourry.net/
> > 
> > Not sure which one we want here
> 
> Yeah, I found your proposal just minutes after sending mine.
> 
> I agree this is whacky syntax, but it's a somewhat common Kconfig
> pattern (see NF_CONNTRACK in net/ipv4/netfilter/Kconfig).
> 
> Anyway, my proposal here should force demotion to "m" for CXL_ACPI if
> PMEM is also "m", where your proposal should force a failure path at
> compile time if ACPI=y and PMEM=m. I really don't know if there's any
> preference at all, I also just wanted the kernel build to complete
> successfuly. :)

my proposal allows compilation but fails during probe because the symbol
is IS_REACHABLE()=false, but yeah, i have no idea what we want here.

I somewhat prefer not to auto-convert config settings, as packaging
changes can be painful - but I can investigate whether this will break
us.

~Gregory

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Keith Busch]

On Thu, Mar 05, 2026 at 05:56:06PM -0500, Gregory Price wrote:
> > Anyway, my proposal here should force demotion to "m" for CXL_ACPI if
> > PMEM is also "m", where your proposal should force a failure path at
> > compile time if ACPI=y and PMEM=m. I really don't know if there's any
> > preference at all, I also just wanted the kernel build to complete
> > successfuly. :)
> 
> my proposal allows compilation but fails during probe because the symbol
> is IS_REACHABLE()=false, but yeah, i have no idea what we want here.

Indeed, I didn't mean to suggest your solution may not build. I only
mean the forced runtime error path was decided at compile time. And I'm
totally okay with doing that if it's preferred over forcing module
demotion.

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[dan.j.williams]

Keith Busch wrote:
> From: Keith Busch <kbusch@kernel.org>
> 
> Commit e7e222ad73d9 moves devm_cxl_add_nvdimm_bridge() into the cxl_pmem
> file, which has independent config compile options for built-in or
> module. The call from cxl_acpi_probe() is guarded by
> IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.
> 
> When CONFIG_CXL_PMEM=m, a built-in cxl_acpi attempts to reference a
> symbol exported by a module, which fails to link. CXL_PMEM cannot simply
> be promoted to =y in this configuration because it depends on LIBNVDIMM,
> which may itself be =m.
> 
> Add a Kconfig dependency to prevent CXL_ACPI from being built-in when
> CXL_PMEM is a module. This contrains CXL_ACPI to =m when CXL_PMEM=m,
> while still allowing CXL_ACPI to be freely configured when CXL_PMEM is
> either built-in or disabled.
> 
> Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> Signed-off-by: Keith Busch <kbusch@kernel.org>
> ---
>  drivers/cxl/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
> index 4589bf11d3fe0..80aeb0d556bd7 100644
> --- a/drivers/cxl/Kconfig
> +++ b/drivers/cxl/Kconfig
> @@ -59,6 +59,7 @@ config CXL_ACPI
>  	tristate "CXL ACPI: Platform Support"
>  	depends on ACPI
>  	depends on ACPI_NUMA
> +	depends on CXL_PMEM || !CXL_PMEM

So I do not think this is sufficient because there is no
devm_cxl_add_nvdimm_bridge() stub in the CONFIG_CXL_PMEM=n case, right?

I came up with this. I do not like that it puts ifdefs in the code, but
then again I do not think add_root_nvdimm_bridge() is big enough to
deserve its own C file.

-- 8< --
diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
index 4589bf11d3fe..277877da8263 100644
--- a/drivers/cxl/Kconfig
+++ b/drivers/cxl/Kconfig
@@ -75,6 +75,11 @@ config CXL_ACPI
 
 	  If unsure say 'm'.
 
+config CXL_ACPI_PMEM
+	def_bool y
+	depends on CXL_ACPI
+	depends on CXL_PMEM >= CXL_ACPI
+
 config CXL_PMEM
 	tristate "CXL PMEM: Persistent Memory Support"
 	depends on LIBNVDIMM
diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index 127537628817..a42ad5035958 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -741,6 +741,7 @@ static int add_host_bridge_uport(struct device *match, void *arg)
 	return 0;
 }
 
+#ifdef CONFIG_CXL_ACPI_PMEM
 static int add_root_nvdimm_bridge(struct device *match, void *data)
 {
 	struct cxl_decoder *cxld;
@@ -765,6 +766,18 @@ static int add_root_nvdimm_bridge(struct device *match, void *data)
 	return 1;
 }
 
+static int setup_pmem(struct cxl_port *root_port)
+{
+	return device_for_each_child(&root_port->dev, root_port,
+				     add_root_nvdimm_bridge);
+}
+#else
+static int setup_pmem(struct cxl_port *root_port)
+{
+	return 0;
+}
+#endif
+
 static struct lock_class_key cxl_root_key;
 
 static void cxl_acpi_lock_reset_class(void *dev)
@@ -952,9 +965,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
 	if (rc < 0)
 		return rc;
 
-	if (IS_ENABLED(CONFIG_CXL_PMEM))
-		rc = device_for_each_child(&root_port->dev, root_port,
-					   add_root_nvdimm_bridge);
+	rc = setup_pmem(root_port);
 	if (rc < 0)
 		return rc;
 

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Keith Busch]

On Thu, Mar 05, 2026 at 05:07:28PM -0800, dan.j.williams@intel.com wrote:
> > The call from cxl_acpi_probe() is guarded by
> > IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.

<snip>

> So I do not think this is sufficient because there is no
> devm_cxl_add_nvdimm_bridge() stub in the CONFIG_CXL_PMEM=n case, right?

It's called from a preprocesser "#if 0" section, so it shouldn't be
compiled in. At least it seems perfectly fine in local testing, but is
that compiler specific? I think this should have been optimized out.

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[dan.j.williams]

Keith Busch wrote:
> On Thu, Mar 05, 2026 at 05:07:28PM -0800, dan.j.williams@intel.com wrote:
> > > The call from cxl_acpi_probe() is guarded by
> > > IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.
> 
> <snip>
> 
> > So I do not think this is sufficient because there is no
> > devm_cxl_add_nvdimm_bridge() stub in the CONFIG_CXL_PMEM=n case, right?
> 
> It's called from a preprocesser "#if 0" section, so it shouldn't be
> compiled in. At least it seems perfectly fine in local testing, but is
> that compiler specific? I think this should have been optimized out.

Oh, duh, yes, the IS_ENABLED() stays there. Hmm, this indeed looks like
it does everything right. I forgot that the dependency checker will make
sure that you can not select CXL_ACPI=y and CXL_PMEM=m.

So,

Reviewed-by: Dan Williams <dan.j.williams@intel.com>

...this is better than what I proposed.

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Jonathan Cameron]

On Thu, 5 Mar 2026 17:55:43 -0800
dan.j.williams@intel.com wrote:

> Keith Busch wrote:
> > On Thu, Mar 05, 2026 at 05:07:28PM -0800, dan.j.williams@intel.com wrote:  
> > > > The call from cxl_acpi_probe() is guarded by
> > > > IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.  
> > 
> > <snip>
> >   
> > > So I do not think this is sufficient because there is no
> > > devm_cxl_add_nvdimm_bridge() stub in the CONFIG_CXL_PMEM=n case, right?  
> > 
> > It's called from a preprocesser "#if 0" section, so it shouldn't be
> > compiled in. At least it seems perfectly fine in local testing, but is
> > that compiler specific? I think this should have been optimized out.  
> 
> Oh, duh, yes, the IS_ENABLED() stays there. Hmm, this indeed looks like
> it does everything right. I forgot that the dependency checker will make
> sure that you can not select CXL_ACPI=y and CXL_PMEM=m.
> 
> So,
> 
> Reviewed-by: Dan Williams <dan.j.williams@intel.com>
> 
> ...this is better than what I proposed.
> 
I prefer Keith's solution to Gregory's just because of removing a surprising foot gun.

Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>

Re: [PATCH] cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch

[Dave Jiang]

On 3/5/26 1:40 PM, Keith Busch wrote:
> From: Keith Busch <kbusch@kernel.org>
> 
> Commit e7e222ad73d9 moves devm_cxl_add_nvdimm_bridge() into the cxl_pmem
> file, which has independent config compile options for built-in or
> module. The call from cxl_acpi_probe() is guarded by
> IS_ENABLED(CONFIG_CXL_PMEM), which evaluates to true for both =y and =m.
> 
> When CONFIG_CXL_PMEM=m, a built-in cxl_acpi attempts to reference a
> symbol exported by a module, which fails to link. CXL_PMEM cannot simply
> be promoted to =y in this configuration because it depends on LIBNVDIMM,
> which may itself be =m.
> 
> Add a Kconfig dependency to prevent CXL_ACPI from being built-in when
> CXL_PMEM is a module. This contrains CXL_ACPI to =m when CXL_PMEM=m,
> while still allowing CXL_ACPI to be freely configured when CXL_PMEM is
> either built-in or disabled.
> 
> Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> Signed-off-by: Keith Busch <kbusch@kernel.org>

Applied to cxl/fixes
93d0fcdddc9e7be9d4f42acbe57bc90dbb0fe75d

> ---
>  drivers/cxl/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
> index 4589bf11d3fe0..80aeb0d556bd7 100644
> --- a/drivers/cxl/Kconfig
> +++ b/drivers/cxl/Kconfig
> @@ -59,6 +59,7 @@ config CXL_ACPI
>  	tristate "CXL ACPI: Platform Support"
>  	depends on ACPI
>  	depends on ACPI_NUMA
> +	depends on CXL_PMEM || !CXL_PMEM
>  	default CXL_BUS
>  	select ACPI_TABLE_LIB
>  	select ACPI_HMAT