1. Patchew
  2. linux
  3. [PATCH 0/6] dax/hmem, cxl: Coordinate Soft Reserved handling with CXL
  4. View patch

[RFC PATCH 6/6] cxl/region, dax/hmem: Guard CXL DAX region creation and tighten HMEM deps

Smita Koralahalli posted 6 patches 1 month, 1 week ago
There is a newer version of this series
[RFC PATCH 6/6] cxl/region, dax/hmem: Guard CXL DAX region creation and tighten HMEM deps
Posted by Smita Koralahalli 1 month, 1 week ago 
Prevent cxl_region_probe() from unconditionally calling into
devm_cxl_add_dax_region() when the DEV_DAX_CXL driver is not enabled.
Wrap the call with IS_ENABLED(CONFIG_DEV_DAX_CXL) so region probe skips
DAX setup cleanly if no consumer is present.

In parallel, update DEV_DAX_HMEM’s Kconfig to depend on
!CXL_BUS || (CXL_ACPI && CXL_PCI) || m. This ensures:

Built-in (y) HMEM is allowed when CXL is disabled, or when the full
CXL discovery stack is built-in. Module (m) HMEM remains always possible.

Signed-off-by: Smita Koralahalli <Smita.KoralahalliChannabasappa@amd.com>
---
I did not want to override Dan’s original approach, so I am posting this
as an RFC.

This patch addresses a corner case when applied on top of Patches 1–5.

When DEV_DAX_HMEM=y and CXL=m, the DEV_DAX_CXL option ends up disabled.
In that configuration, with Patches 1–5 applied, ownership of the Soft
Reserved ranges falls back to dax_hmem. As a result, /proc/iomem looks
like this:

850000000-284fffffff : CXL Window 0
  850000000-284fffffff : region3
    850000000-284fffffff : Soft Reserved
      850000000-284fffffff : dax0.0
        850000000-284fffffff : System RAM (kmem)
2850000000-484fffffff : CXL Window 1
  2850000000-484fffffff : region4
    2850000000-484fffffff : Soft Reserved
      2850000000-484fffffff : dax1.0
        2850000000-484fffffff : System RAM (kmem)
4850000000-684fffffff : CXL Window 2
  4850000000-684fffffff : region5
    4850000000-684fffffff : Soft Reserved
      4850000000-684fffffff : dax2.0
        4850000000-684fffffff : System RAM (kmem)

In this case the dax devices are created by dax_hmem, not by dax_cxl.
Consequently, a "cxl disable-region <regionx>" operation does not
unregister these devices. In addition, the dmesg output can be misleading
to users, since it looks like the CXL region driver created the devdax
devices:

  devm_cxl_add_region: cxl_acpi ACPI0017:00: decoder0.2: created region5
  ..
  ..

This patch addresses those situations. I am not entirely sure how clean
the approach of using “|| m” is, so I am sending it as RFC for feedback.
---
 drivers/cxl/core/region.c | 4 +++-
 drivers/dax/Kconfig       | 1 +
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
index 71cc42d05248..6a2c21e55dbc 100644
--- a/drivers/cxl/core/region.c
+++ b/drivers/cxl/core/region.c
@@ -3617,7 +3617,9 @@ static int cxl_region_probe(struct device *dev)
 					p->res->start, p->res->end, cxlr,
 					is_system_ram) > 0)
 			return 0;
-		return devm_cxl_add_dax_region(cxlr);
+		if (IS_ENABLED(CONFIG_DEV_DAX_CXL))
+			return devm_cxl_add_dax_region(cxlr);
+		return 0;
 	default:
 		dev_dbg(&cxlr->dev, "unsupported region mode: %d\n",
 			cxlr->mode);
diff --git a/drivers/dax/Kconfig b/drivers/dax/Kconfig
index 3683bb3f2311..fd12cca91c78 100644
--- a/drivers/dax/Kconfig
+++ b/drivers/dax/Kconfig
@@ -30,6 +30,7 @@ config DEV_DAX_PMEM
 config DEV_DAX_HMEM
 	tristate "HMEM DAX: direct access to 'specific purpose' memory"
 	depends on EFI_SOFT_RESERVE
+	depends on !CXL_BUS || (CXL_ACPI && CXL_PCI) || m
 	select NUMA_KEEP_MEMINFO if NUMA_MEMBLKS
 	default DEV_DAX
 	help
-- 
2.17.1
Re: [RFC PATCH 6/6] cxl/region, dax/hmem: Guard CXL DAX region creation and tighten HMEM deps
Posted by Zhijian Li (Fujitsu) 1 month ago 

On 22/08/2025 11:42, Smita Koralahalli wrote:
> Prevent cxl_region_probe() from unconditionally calling into
> devm_cxl_add_dax_region() when the DEV_DAX_CXL driver is not enabled.
> Wrap the call with IS_ENABLED(CONFIG_DEV_DAX_CXL) so region probe skips
> DAX setup cleanly if no consumer is present.

A question came to mind:
  
Why is the case of `CXL_REGION && !DEV_DAX_CXL` necessary? It appears to fall back to the hmem driver in that scenario.
If so, could we instead simplify it as follows?
  
--- a/drivers/cxl/Kconfig
+++ b/drivers/cxl/Kconfig
@@ -200,6 +200,7 @@ config CXL_REGION
         depends on SPARSEMEM
         select MEMREGION
         select GET_FREE_REGION
+       select DEV_DAX_CXL

> 
> In parallel, update DEV_DAX_HMEM’s Kconfig to depend on
> !CXL_BUS || (CXL_ACPI && CXL_PCI) || m. This ensures:
> 
> Built-in (y) HMEM is allowed when CXL is disabled, or when the full
> CXL discovery stack is built-in. Module (m) HMEM remains always possible.

Hmm,IIUC, `dax_hmem` isn't exclusively designed for CXL. It could support other special memory types (e.g., HBM).

Thanks
Zhijian



> 
> Signed-off-by: Smita Koralahalli <Smita.KoralahalliChannabasappa@amd.com>
> ---
> I did not want to override Dan’s original approach, so I am posting this
> as an RFC.
> 
> This patch addresses a corner case when applied on top of Patches 1–5.
> 
> When DEV_DAX_HMEM=y and CXL=m, the DEV_DAX_CXL option ends up disabled.
> In that configuration, with Patches 1–5 applied, ownership of the Soft
> Reserved ranges falls back to dax_hmem. As a result, /proc/iomem looks
> like this:
> 
> 850000000-284fffffff : CXL Window 0
>    850000000-284fffffff : region3
>      850000000-284fffffff : Soft Reserved
>        850000000-284fffffff : dax0.0
>          850000000-284fffffff : System RAM (kmem)
> 2850000000-484fffffff : CXL Window 1
>    2850000000-484fffffff : region4
>      2850000000-484fffffff : Soft Reserved
>        2850000000-484fffffff : dax1.0
>          2850000000-484fffffff : System RAM (kmem)
> 4850000000-684fffffff : CXL Window 2
>    4850000000-684fffffff : region5
>      4850000000-684fffffff : Soft Reserved
>        4850000000-684fffffff : dax2.0
>          4850000000-684fffffff : System RAM (kmem)
> 
> In this case the dax devices are created by dax_hmem, not by dax_cxl.
> Consequently, a "cxl disable-region <regionx>" operation does not
> unregister these devices. In addition, the dmesg output can be misleading
> to users, since it looks like the CXL region driver created the devdax
> devices:
> 
>    devm_cxl_add_region: cxl_acpi ACPI0017:00: decoder0.2: created region5
>    ..
>    ..
> 
> This patch addresses those situations. I am not entirely sure how clean
> the approach of using “|| m” is, so I am sending it as RFC for feedback.
> ---
>   drivers/cxl/core/region.c | 4 +++-
>   drivers/dax/Kconfig       | 1 +
>   2 files changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index 71cc42d05248..6a2c21e55dbc 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
> @@ -3617,7 +3617,9 @@ static int cxl_region_probe(struct device *dev)
>   					p->res->start, p->res->end, cxlr,
>   					is_system_ram) > 0)
>   			return 0;
> -		return devm_cxl_add_dax_region(cxlr);
> +		if (IS_ENABLED(CONFIG_DEV_DAX_CXL))
> +			return devm_cxl_add_dax_region(cxlr);
> +		return 0;
>   	default:
>   		dev_dbg(&cxlr->dev, "unsupported region mode: %d\n",
>   			cxlr->mode);
> diff --git a/drivers/dax/Kconfig b/drivers/dax/Kconfig
> index 3683bb3f2311..fd12cca91c78 100644
> --- a/drivers/dax/Kconfig
> +++ b/drivers/dax/Kconfig
> @@ -30,6 +30,7 @@ config DEV_DAX_PMEM
>   config DEV_DAX_HMEM
>   	tristate "HMEM DAX: direct access to 'specific purpose' memory"
>   	depends on EFI_SOFT_RESERVE
> +	depends on !CXL_BUS || (CXL_ACPI && CXL_PCI) || m
>   	select NUMA_KEEP_MEMINFO if NUMA_MEMBLKS
>   	default DEV_DAX
>   	help
Re: [RFC PATCH 6/6] cxl/region, dax/hmem: Guard CXL DAX region creation and tighten HMEM deps
Posted by Koralahalli Channabasappa, Smita 3 days, 20 hours ago 
On 8/31/2025 11:21 PM, Zhijian Li (Fujitsu) wrote:
> 
> 
> On 22/08/2025 11:42, Smita Koralahalli wrote:
>> Prevent cxl_region_probe() from unconditionally calling into
>> devm_cxl_add_dax_region() when the DEV_DAX_CXL driver is not enabled.
>> Wrap the call with IS_ENABLED(CONFIG_DEV_DAX_CXL) so region probe skips
>> DAX setup cleanly if no consumer is present.
> 
> A question came to mind:
>    
> Why is the case of `CXL_REGION && !DEV_DAX_CXL` necessary? It appears to fall back to the hmem driver in that scenario.
> If so, could we instead simplify it as follows?
>    
> --- a/drivers/cxl/Kconfig
> +++ b/drivers/cxl/Kconfig
> @@ -200,6 +200,7 @@ config CXL_REGION
>           depends on SPARSEMEM
>           select MEMREGION
>           select GET_FREE_REGION
> +       select DEV_DAX_CXL
> 

I’m not entirely sure about the full implications of disabling 
CXL_REGION when DEV_DAX_CXL is disabled.

The primary intent of this patch was to address the scenario where 
DEV_DAX_HMEM=y and CXL=m, which results in DEV_DAX_CXL being disabled. 
In that configuration, ownership of the soft-reserved ranges incorrectly 
falls back to hmem instead of being managed by CXL. This leads to 
misleading output in /proc/iomem, as I illustrated earlier.

That said, as you pointed out, dax_hmem is not exclusive to CXL, so I 
will drop this patch in v2. The next revision will therefore not cover 
the case of DEV_DAX_HMEM=y and CXL=m. I would appreciate input on how 
best to handle this scenario efficiently.

Thanks
Smita

>>
>> In parallel, update DEV_DAX_HMEM’s Kconfig to depend on
>> !CXL_BUS || (CXL_ACPI && CXL_PCI) || m. This ensures:
>>
>> Built-in (y) HMEM is allowed when CXL is disabled, or when the full
>> CXL discovery stack is built-in. Module (m) HMEM remains always possible.
> 
> Hmm,IIUC, `dax_hmem` isn't exclusively designed for CXL. It could support other special memory types (e.g., HBM).
> 
> Thanks
> Zhijian
> 
> 
> 
>>
>> Signed-off-by: Smita Koralahalli <Smita.KoralahalliChannabasappa@amd.com>
>> ---
>> I did not want to override Dan’s original approach, so I am posting this
>> as an RFC.
>>
>> This patch addresses a corner case when applied on top of Patches 1–5.
>>
>> When DEV_DAX_HMEM=y and CXL=m, the DEV_DAX_CXL option ends up disabled.
>> In that configuration, with Patches 1–5 applied, ownership of the Soft
>> Reserved ranges falls back to dax_hmem. As a result, /proc/iomem looks
>> like this:
>>
>> 850000000-284fffffff : CXL Window 0
>>     850000000-284fffffff : region3
>>       850000000-284fffffff : Soft Reserved
>>         850000000-284fffffff : dax0.0
>>           850000000-284fffffff : System RAM (kmem)
>> 2850000000-484fffffff : CXL Window 1
>>     2850000000-484fffffff : region4
>>       2850000000-484fffffff : Soft Reserved
>>         2850000000-484fffffff : dax1.0
>>           2850000000-484fffffff : System RAM (kmem)
>> 4850000000-684fffffff : CXL Window 2
>>     4850000000-684fffffff : region5
>>       4850000000-684fffffff : Soft Reserved
>>         4850000000-684fffffff : dax2.0
>>           4850000000-684fffffff : System RAM (kmem)
>>
>> In this case the dax devices are created by dax_hmem, not by dax_cxl.
>> Consequently, a "cxl disable-region <regionx>" operation does not
>> unregister these devices. In addition, the dmesg output can be misleading
>> to users, since it looks like the CXL region driver created the devdax
>> devices:
>>
>>     devm_cxl_add_region: cxl_acpi ACPI0017:00: decoder0.2: created region5
>>     ..
>>     ..
>>
>> This patch addresses those situations. I am not entirely sure how clean
>> the approach of using “|| m” is, so I am sending it as RFC for feedback.
>> ---
>>    drivers/cxl/core/region.c | 4 +++-
>>    drivers/dax/Kconfig       | 1 +
>>    2 files changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
>> index 71cc42d05248..6a2c21e55dbc 100644
>> --- a/drivers/cxl/core/region.c
>> +++ b/drivers/cxl/core/region.c
>> @@ -3617,7 +3617,9 @@ static int cxl_region_probe(struct device *dev)
>>    					p->res->start, p->res->end, cxlr,
>>    					is_system_ram) > 0)
>>    			return 0;
>> -		return devm_cxl_add_dax_region(cxlr);
>> +		if (IS_ENABLED(CONFIG_DEV_DAX_CXL))
>> +			return devm_cxl_add_dax_region(cxlr);
>> +		return 0;
>>    	default:
>>    		dev_dbg(&cxlr->dev, "unsupported region mode: %d\n",
>>    			cxlr->mode);
>> diff --git a/drivers/dax/Kconfig b/drivers/dax/Kconfig
>> index 3683bb3f2311..fd12cca91c78 100644
>> --- a/drivers/dax/Kconfig
>> +++ b/drivers/dax/Kconfig
>> @@ -30,6 +30,7 @@ config DEV_DAX_PMEM
>>    config DEV_DAX_HMEM
>>    	tristate "HMEM DAX: direct access to 'specific purpose' memory"
>>    	depends on EFI_SOFT_RESERVE
>> +	depends on !CXL_BUS || (CXL_ACPI && CXL_PCI) || m
>>    	select NUMA_KEEP_MEMINFO if NUMA_MEMBLKS
>>    	default DEV_DAX
>>    	help

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.