[PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[Gregory Price]

Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.

When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
link against a non-built-in cxl_pmem, which the linker cannot resolve.

CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
at most =m.

Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
built-in caller references a modular target, preventing the link error.

The result is if the pmem/nvdimm symbols are not reachable at build
time, then at runtime it will always return -ENODEV.

Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
Signed-off-by: Gregory Price <gourry@gourry.net>
---
 drivers/cxl/acpi.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index 127537628817..7065413eda9f 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
 	if (rc < 0)
 		return rc;
 
-	if (IS_ENABLED(CONFIG_CXL_PMEM))
+	if (IS_REACHABLE(CONFIG_CXL_PMEM))
 		rc = device_for_each_child(&root_port->dev, root_port,
 					   add_root_nvdimm_bridge);
 	if (rc < 0)
-- 
2.53.0

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[Harshit Mogalapalli]

Hi Gregory,

On 03/03/26 01:34, Gregory Price wrote:
> Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> 
> When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> link against a non-built-in cxl_pmem, which the linker cannot resolve.
> 
> CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> at most =m.
> 
> Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> built-in caller references a modular target, preventing the link error.
> 
> The result is if the pmem/nvdimm symbols are not reachable at build
> time, then at runtime it will always return -ENODEV.
> 
> Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> Signed-off-by: Gregory Price <gourry@gourry.net>

We have run into the same issue as we have CXL_PMEM=m and CXL_ACPI=y, 
and this fix looks good to me.

Reviewed-by: Harshit Mogalapalli <harshit.m.mogalapalli@oracle.com>

also, gave this patch a go with our config:

Tested-by: Harshit Mogalapalli <harshit.m.mogalapalli@oracle.com>



Thanks,
Harshit

> ---
>   drivers/cxl/acpi.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> index 127537628817..7065413eda9f 100644
> --- a/drivers/cxl/acpi.c
> +++ b/drivers/cxl/acpi.c
> @@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
>   	if (rc < 0)
>   		return rc;
>   
> -	if (IS_ENABLED(CONFIG_CXL_PMEM))
> +	if (IS_REACHABLE(CONFIG_CXL_PMEM))
>   		rc = device_for_each_child(&root_port->dev, root_port,
>   					   add_root_nvdimm_bridge);
>   	if (rc < 0)

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[Jonathan Cameron]

On Mon,  2 Mar 2026 15:04:29 -0500
Gregory Price <gourry@gourry.net> wrote:

> Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> 
> When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> link against a non-built-in cxl_pmem, which the linker cannot resolve.
> 
> CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> at most =m.
> 
> Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> built-in caller references a modular target, preventing the link error.
> 
> The result is if the pmem/nvdimm symbols are not reachable at build
> time, then at runtime it will always return -ENODEV.
> 
> Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> Signed-off-by: Gregory Price <gourry@gourry.net>

What's the impact of this build combination wrt to whether the result is
actually useful?   Is there any point in building CXL_PMEM as a module
without this call being made?  I can't remember and normally don't care
much about PMEM :) 

> ---
>  drivers/cxl/acpi.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> index 127537628817..7065413eda9f 100644
> --- a/drivers/cxl/acpi.c
> +++ b/drivers/cxl/acpi.c
> @@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
>  	if (rc < 0)
>  		return rc;
>  
> -	if (IS_ENABLED(CONFIG_CXL_PMEM))
> +	if (IS_REACHABLE(CONFIG_CXL_PMEM))
>  		rc = device_for_each_child(&root_port->dev, root_port,
>  					   add_root_nvdimm_bridge);
>  	if (rc < 0)

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[Gregory Price]

On Tue, Mar 03, 2026 at 10:13:30AM +0000, Jonathan Cameron wrote:
> On Mon,  2 Mar 2026 15:04:29 -0500
> Gregory Price <gourry@gourry.net> wrote:
> 
> > Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> > cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> > cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> > 
> > When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> > link against a non-built-in cxl_pmem, which the linker cannot resolve.
> > 
> > CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> > at most =m.
> > 
> > Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> > built-in caller references a modular target, preventing the link error.
> > 
> > The result is if the pmem/nvdimm symbols are not reachable at build
> > time, then at runtime it will always return -ENODEV.
> > 
> > Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> > Signed-off-by: Gregory Price <gourry@gourry.net>
> 
> What's the impact of this build combination wrt to whether the result is
> actually useful?   Is there any point in building CXL_PMEM as a module
> without this call being made?  I can't remember and normally don't care
> much about PMEM :) 
> 

Basically no, it's not useful, but it's some definition of the default
depending on your olddefconfig, so rather than fight with this further
this basically terminates the same way DAX_CXL=m terminates.

tl;dr: You don't create the dax device / nbdimm and you're sad... but it
does build!

~Gregory

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[dan.j.williams]

Gregory Price wrote:
> Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> 
> When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> link against a non-built-in cxl_pmem, which the linker cannot resolve.
> 
> CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> at most =m.
> 
> Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> built-in caller references a modular target, preventing the link error.
> 
> The result is if the pmem/nvdimm symbols are not reachable at build
> time, then at runtime it will always return -ENODEV.
> 
> Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> Signed-off-by: Gregory Price <gourry@gourry.net>
> ---
>  drivers/cxl/acpi.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> index 127537628817..7065413eda9f 100644
> --- a/drivers/cxl/acpi.c
> +++ b/drivers/cxl/acpi.c
> @@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
>  	if (rc < 0)
>  		return rc;
>  
> -	if (IS_ENABLED(CONFIG_CXL_PMEM))
> +	if (IS_REACHABLE(CONFIG_CXL_PMEM))
>  		rc = device_for_each_child(&root_port->dev, root_port,
>  					   add_root_nvdimm_bridge);

I think we want this conflict to be resolved at build time. The nice
thing about a new CONFIG_CXL_ACPI_PMEM symbol is you can check your
config to find out why your CXL PMEM is not working. Otherwise,
IS_REACHABLE() hides this subtelty.

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[Song Liu]

On Thu, Mar 5, 2026 at 5:23 PM <dan.j.williams@intel.com> wrote:
>
> Gregory Price wrote:
> > Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> > cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> > cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> >
> > When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> > link against a non-built-in cxl_pmem, which the linker cannot resolve.
> >
> > CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> > at most =m.
> >
> > Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> > built-in caller references a modular target, preventing the link error.
> >
> > The result is if the pmem/nvdimm symbols are not reachable at build
> > time, then at runtime it will always return -ENODEV.
> >
> > Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> > Signed-off-by: Gregory Price <gourry@gourry.net>
> > ---
> >  drivers/cxl/acpi.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > index 127537628817..7065413eda9f 100644
> > --- a/drivers/cxl/acpi.c
> > +++ b/drivers/cxl/acpi.c
> > @@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
> >       if (rc < 0)
> >               return rc;
> >
> > -     if (IS_ENABLED(CONFIG_CXL_PMEM))
> > +     if (IS_REACHABLE(CONFIG_CXL_PMEM))
> >               rc = device_for_each_child(&root_port->dev, root_port,
> >                                          add_root_nvdimm_bridge);
>
> I think we want this conflict to be resolved at build time. The nice
> thing about a new CONFIG_CXL_ACPI_PMEM symbol is you can check your
> config to find out why your CXL PMEM is not working. Otherwise,
> IS_REACHABLE() hides this subtelty.

Can we fix this with something like the following?

Thanks,
Song

diff --git i/drivers/cxl/Kconfig w/drivers/cxl/Kconfig
index 4589bf11d3fe..0290e0f22cf8 100644
--- i/drivers/cxl/Kconfig
+++ w/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
        default CXL_BUS
        select ACPI_TABLE_LIB
        select ACPI_HMAT

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[dan.j.williams]

Song Liu wrote:
> On Thu, Mar 5, 2026 at 5:23 PM <dan.j.williams@intel.com> wrote:
> >
> > Gregory Price wrote:
> > > Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> > > cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> > > cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> > >
> > > When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> > > link against a non-built-in cxl_pmem, which the linker cannot resolve.
> > >
> > > CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> > > at most =m.
> > >
> > > Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> > > built-in caller references a modular target, preventing the link error.
> > >
> > > The result is if the pmem/nvdimm symbols are not reachable at build
> > > time, then at runtime it will always return -ENODEV.
> > >
> > > Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> > > Signed-off-by: Gregory Price <gourry@gourry.net>
> > > ---
> > >  drivers/cxl/acpi.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > > index 127537628817..7065413eda9f 100644
> > > --- a/drivers/cxl/acpi.c
> > > +++ b/drivers/cxl/acpi.c
> > > @@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
> > >       if (rc < 0)
> > >               return rc;
> > >
> > > -     if (IS_ENABLED(CONFIG_CXL_PMEM))
> > > +     if (IS_REACHABLE(CONFIG_CXL_PMEM))
> > >               rc = device_for_each_child(&root_port->dev, root_port,
> > >                                          add_root_nvdimm_bridge);
> >
> > I think we want this conflict to be resolved at build time. The nice
> > thing about a new CONFIG_CXL_ACPI_PMEM symbol is you can check your
> > config to find out why your CXL PMEM is not working. Otherwise,
> > IS_REACHABLE() hides this subtelty.
> 
> Can we fix this with something like the following?

BTW, this is queued now:

https://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl.git/commit/?h=fixes&id=93d0fcdddc9e

Re: [PATCH] cxl/acpi: fix CXL_ACPI=y + CXL_PMEM=m link failure

[Song Liu]

On Fri, Mar 6, 2026 at 5:33 PM <dan.j.williams@intel.com> wrote:
>
> Song Liu wrote:
> > On Thu, Mar 5, 2026 at 5:23 PM <dan.j.williams@intel.com> wrote:
> > >
> > > Gregory Price wrote:
> > > > Commit e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to
> > > > cxl_pmem.ko") moved devm_cxl_add_nvdimm_bridge() from cxl_core into
> > > > cxl_pmem, creating a symbol dependency from cxl_acpi to cxl_pmem.
> > > >
> > > > When CXL_ACPI=y and CXL_PMEM=m, the built-in cxl_acpi attempts to
> > > > link against a non-built-in cxl_pmem, which the linker cannot resolve.
> > > >
> > > > CXL_PMEM depends on LIBNVDIMM, so LIBNVDIMM=m constrains CXL_PMEM to
> > > > at most =m.
> > > >
> > > > Change IS_ENABLED() to IS_REACHABLE(), which returns false when a
> > > > built-in caller references a modular target, preventing the link error.
> > > >
> > > > The result is if the pmem/nvdimm symbols are not reachable at build
> > > > time, then at runtime it will always return -ENODEV.
> > > >
> > > > Fixes: e7e222ad73d9 ("cxl: Move devm_cxl_add_nvdimm_bridge() to cxl_pmem.ko")
> > > > Signed-off-by: Gregory Price <gourry@gourry.net>
> > > > ---
> > > >  drivers/cxl/acpi.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > > > index 127537628817..7065413eda9f 100644
> > > > --- a/drivers/cxl/acpi.c
> > > > +++ b/drivers/cxl/acpi.c
> > > > @@ -952,7 +952,7 @@ static int cxl_acpi_probe(struct platform_device *pdev)
> > > >       if (rc < 0)
> > > >               return rc;
> > > >
> > > > -     if (IS_ENABLED(CONFIG_CXL_PMEM))
> > > > +     if (IS_REACHABLE(CONFIG_CXL_PMEM))
> > > >               rc = device_for_each_child(&root_port->dev, root_port,
> > > >                                          add_root_nvdimm_bridge);
> > >
> > > I think we want this conflict to be resolved at build time. The nice
> > > thing about a new CONFIG_CXL_ACPI_PMEM symbol is you can check your
> > > config to find out why your CXL PMEM is not working. Otherwise,
> > > IS_REACHABLE() hides this subtelty.
> >
> > Can we fix this with something like the following?
>
> BTW, this is queued now:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl.git/commit/?h=fixes&id=93d0fcdddc9e

Aha, missed this one. Thanks for the information.

Song