1. Patchew
  2. linux
  3. View series

[PATCH v2] cxl/test: Enforce PMD alignment for volatile mock regions

Richard Cheng posted 1 patch 3 days, 12 hours ago
There is a newer version of this series
tools/testing/cxl/test/cxl.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
[PATCH v2] cxl/test: Enforce PMD alignment for volatile mock regions
Posted by Richard Cheng 3 days, 12 hours ago 
cxl_test allocates synthetic CFMWS HPA windows from a gen_pool with
SZ_256M alignment. On arm64 with CONFIG_ARM64_64K_PAGES=y and
CONFIG_PGTABLE_LEVELS=3, PMD_SIZE is 512M, so every CXL region carved
from a volatile window inherits a non-PMD-aligned start, and
cxl_dax_region_probe() -> alloca_dax_region() fails:

"""
    cxl_dax_region dax_region1: probe with driver cxl_dax_region failed
                                with error -12
"""

Enfore that every volatile mock CFMWS is PMD-aligned in both start and
size.

Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Acked-by: Kai-Heng Feng <kaihengf@nvidia.com>
Signed-off-by: Richard Cheng <icheng@nvidia.com>
---
Changelog:

v1 -> v2:
    - Reframe from "fix start alignment" to "enforce PMD invariant for
      volatile CFMWS"
    - Bump cfmws5.window_size to max(SZ_256M, PMD_SIZE)
    - Gate populate_cedt() alignment on
      ACPI_CEDT_CFMWS_RESTRICT_VOLATILE so PMEM windows stay at SZ_256M
    - Add BUILD_BUG_ON for MOCK_AUTO_REGION_SIZE_DEFAULT v.s. PMD_SIZE
      in cxl_test_init()

v1 link: https://lore.kernel.org/r/20260519233523.5991-1-icheng@nvidia.com/

Best regards,
Richard Cheng.
---
 tools/testing/cxl/test/cxl.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
index 418669927fb0..09c309af5fd1 100644
--- a/tools/testing/cxl/test/cxl.c
+++ b/tools/testing/cxl/test/cxl.c
@@ -318,7 +318,7 @@ static struct {
 			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
 					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE,
 			.qtg_id = FAKE_QTG_ID,
-			.window_size = SZ_256M,
+			.window_size = SZ_256M > PMD_SIZE ? SZ_256M : PMD_SIZE,
 		},
 		.target = { 3 },
 	},
@@ -495,9 +495,17 @@ static int populate_cedt(void)
 
 	for (i = cfmws_start; i <= cfmws_end; i++) {
 		struct acpi_cedt_cfmws *window = mock_cfmws[i];
+		int align = SZ_256M;
 
 		cfmws_elc_update(window, i);
-		res = alloc_mock_res(window->window_size, SZ_256M);
+		if (window->restrictions & ACPI_CEDT_CFMWS_RESTRICT_VOLATILE)
+			align = max_t(int, SZ_256M, PMD_SIZE);
+		res = alloc_mock_res(window->window_size, align);
 		if (!res)
 			return -ENOMEM;
 		window->base_hpa = res->range.start;
@@ -1826,6 +1834,8 @@ static __init int cxl_test_init(void)
 	int rc, i;
 	struct range mappable;
 
+	BUILD_BUG_ON(!IS_ALIGNED(MOCK_AUTO_REGION_SIZE_DEFAULT, PMD_SIZE));
+
 	cxl_acpi_test();
 	cxl_core_test();
 	cxl_mem_test();
-- 
2.43.0
Re: [PATCH v2] cxl/test: Enforce PMD alignment for volatile mock regions
Posted by Dave Jiang 3 days, 3 hours ago 

On 5/21/26 1:32 AM, Richard Cheng wrote:
> cxl_test allocates synthetic CFMWS HPA windows from a gen_pool with
> SZ_256M alignment. On arm64 with CONFIG_ARM64_64K_PAGES=y and
> CONFIG_PGTABLE_LEVELS=3, PMD_SIZE is 512M, so every CXL region carved
> from a volatile window inherits a non-PMD-aligned start, and
> cxl_dax_region_probe() -> alloca_dax_region() fails:
> 
> """
>     cxl_dax_region dax_region1: probe with driver cxl_dax_region failed
>                                 with error -12
> """
> 
> Enfore that every volatile mock CFMWS is PMD-aligned in both start and
> size.
> 
> Reviewed-by: Dave Jiang <dave.jiang@intel.com>
> Acked-by: Kai-Heng Feng <kaihengf@nvidia.com>
> Signed-off-by: Richard Cheng <icheng@nvidia.com>

Hi Richard,
I can't seem to apply this patch against v7.1-rc4. What base did you use?

DJ



> ---
> Changelog:
> 
> v1 -> v2:
>     - Reframe from "fix start alignment" to "enforce PMD invariant for
>       volatile CFMWS"
>     - Bump cfmws5.window_size to max(SZ_256M, PMD_SIZE)
>     - Gate populate_cedt() alignment on
>       ACPI_CEDT_CFMWS_RESTRICT_VOLATILE so PMEM windows stay at SZ_256M
>     - Add BUILD_BUG_ON for MOCK_AUTO_REGION_SIZE_DEFAULT v.s. PMD_SIZE
>       in cxl_test_init()
> 
> v1 link: https://lore.kernel.org/r/20260519233523.5991-1-icheng@nvidia.com/
> 
> Best regards,
> Richard Cheng.
> ---
>  tools/testing/cxl/test/cxl.c | 14 ++++++++++++--
>  1 file changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
> index 418669927fb0..09c309af5fd1 100644
> --- a/tools/testing/cxl/test/cxl.c
> +++ b/tools/testing/cxl/test/cxl.c
> @@ -318,7 +318,7 @@ static struct {
>  			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
>  					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE,
>  			.qtg_id = FAKE_QTG_ID,
> -			.window_size = SZ_256M,
> +			.window_size = SZ_256M > PMD_SIZE ? SZ_256M : PMD_SIZE,
>  		},
>  		.target = { 3 },
>  	},
> @@ -495,9 +495,17 @@ static int populate_cedt(void)
>  
>  	for (i = cfmws_start; i <= cfmws_end; i++) {
>  		struct acpi_cedt_cfmws *window = mock_cfmws[i];
> +		int align = SZ_256M;
>  
>  		cfmws_elc_update(window, i);
> -		res = alloc_mock_res(window->window_size, SZ_256M);
> +		if (window->restrictions & ACPI_CEDT_CFMWS_RESTRICT_VOLATILE)
> +			align = max_t(int, SZ_256M, PMD_SIZE);
> +		res = alloc_mock_res(window->window_size, align);
>  		if (!res)
>  			return -ENOMEM;
>  		window->base_hpa = res->range.start;
> @@ -1826,6 +1834,8 @@ static __init int cxl_test_init(void)
>  	int rc, i;
>  	struct range mappable;
>  
> +	BUILD_BUG_ON(!IS_ALIGNED(MOCK_AUTO_REGION_SIZE_DEFAULT, PMD_SIZE));
> +
>  	cxl_acpi_test();
>  	cxl_core_test();
>  	cxl_mem_test();
Re: [PATCH v2] cxl/test: Enforce PMD alignment for volatile mock regions
Posted by Alison Schofield 3 days ago 
On Thu, May 21, 2026 at 10:20:26AM -0700, Dave Jiang wrote:
> 
> 
> On 5/21/26 1:32 AM, Richard Cheng wrote:
> > cxl_test allocates synthetic CFMWS HPA windows from a gen_pool with
> > SZ_256M alignment. On arm64 with CONFIG_ARM64_64K_PAGES=y and
> > CONFIG_PGTABLE_LEVELS=3, PMD_SIZE is 512M, so every CXL region carved
> > from a volatile window inherits a non-PMD-aligned start, and
> > cxl_dax_region_probe() -> alloca_dax_region() fails:
> > 
> > """
> >     cxl_dax_region dax_region1: probe with driver cxl_dax_region failed
> >                                 with error -12
> > """
> > 
> > Enfore that every volatile mock CFMWS is PMD-aligned in both start and
> > size.
> > 
> > Reviewed-by: Dave Jiang <dave.jiang@intel.com>
> > Acked-by: Kai-Heng Feng <kaihengf@nvidia.com>
> > Signed-off-by: Richard Cheng <icheng@nvidia.com>
> 
> Hi Richard,
> I can't seem to apply this patch against v7.1-rc4. What base did you use?

Hunk headers are wrong. Maybe hand editing without reformatting.
Diffstat is wrong too but that didn't affect application.

I'll wait for the v3.


> 
> DJ
> 
> 
> 
> > ---
> > Changelog:
> > 
> > v1 -> v2:
> >     - Reframe from "fix start alignment" to "enforce PMD invariant for
> >       volatile CFMWS"
> >     - Bump cfmws5.window_size to max(SZ_256M, PMD_SIZE)
> >     - Gate populate_cedt() alignment on
> >       ACPI_CEDT_CFMWS_RESTRICT_VOLATILE so PMEM windows stay at SZ_256M
> >     - Add BUILD_BUG_ON for MOCK_AUTO_REGION_SIZE_DEFAULT v.s. PMD_SIZE
> >       in cxl_test_init()
> > 
> > v1 link: https://lore.kernel.org/r/20260519233523.5991-1-icheng@nvidia.com/
> > 
> > Best regards,
> > Richard Cheng.
> > ---
> >  tools/testing/cxl/test/cxl.c | 14 ++++++++++++--
> >  1 file changed, 12 insertions(+), 2 deletions(-)
> > 
> > diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
> > index 418669927fb0..09c309af5fd1 100644
> > --- a/tools/testing/cxl/test/cxl.c
> > +++ b/tools/testing/cxl/test/cxl.c
> > @@ -318,7 +318,7 @@ static struct {
> >  			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
> >  					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE,
> >  			.qtg_id = FAKE_QTG_ID,
> > -			.window_size = SZ_256M,
> > +			.window_size = SZ_256M > PMD_SIZE ? SZ_256M : PMD_SIZE,
> >  		},
> >  		.target = { 3 },
> >  	},
> > @@ -495,9 +495,17 @@ static int populate_cedt(void)
> >  
> >  	for (i = cfmws_start; i <= cfmws_end; i++) {
> >  		struct acpi_cedt_cfmws *window = mock_cfmws[i];
> > +		int align = SZ_256M;
> >  
> >  		cfmws_elc_update(window, i);
> > -		res = alloc_mock_res(window->window_size, SZ_256M);
> > +		if (window->restrictions & ACPI_CEDT_CFMWS_RESTRICT_VOLATILE)
> > +			align = max_t(int, SZ_256M, PMD_SIZE);
> > +		res = alloc_mock_res(window->window_size, align);
> >  		if (!res)
> >  			return -ENOMEM;
> >  		window->base_hpa = res->range.start;
> > @@ -1826,6 +1834,8 @@ static __init int cxl_test_init(void)
> >  	int rc, i;
> >  	struct range mappable;
> >  
> > +	BUILD_BUG_ON(!IS_ALIGNED(MOCK_AUTO_REGION_SIZE_DEFAULT, PMD_SIZE));
> > +
> >  	cxl_acpi_test();
> >  	cxl_core_test();
> >  	cxl_mem_test();
>
Re: [PATCH v2] cxl/test: Enforce PMD alignment for volatile mock regions
Posted by Richard Cheng 2 days, 14 hours ago 
On Thu, May 21, 2026 at 01:24:49PM +0800, Alison Schofield wrote:
> On Thu, May 21, 2026 at 10:20:26AM -0700, Dave Jiang wrote:
> > 
> > 
> > On 5/21/26 1:32 AM, Richard Cheng wrote:
> > > cxl_test allocates synthetic CFMWS HPA windows from a gen_pool with
> > > SZ_256M alignment. On arm64 with CONFIG_ARM64_64K_PAGES=y and
> > > CONFIG_PGTABLE_LEVELS=3, PMD_SIZE is 512M, so every CXL region carved
> > > from a volatile window inherits a non-PMD-aligned start, and
> > > cxl_dax_region_probe() -> alloca_dax_region() fails:
> > > 
> > > """
> > >     cxl_dax_region dax_region1: probe with driver cxl_dax_region failed
> > >                                 with error -12
> > > """
> > > 
> > > Enfore that every volatile mock CFMWS is PMD-aligned in both start and
> > > size.
> > > 
> > > Reviewed-by: Dave Jiang <dave.jiang@intel.com>
> > > Acked-by: Kai-Heng Feng <kaihengf@nvidia.com>
> > > Signed-off-by: Richard Cheng <icheng@nvidia.com>
> > 
> > Hi Richard,
> > I can't seem to apply this patch against v7.1-rc4. What base did you use?
> 
> Hunk headers are wrong. Maybe hand editing without reformatting.
> Diffstat is wrong too but that didn't affect application.
> 
> I'll wait for the v3.
> 
>
Hi Dave, Alison,

Sorry I hand eddited the sequence of this line
+                       .window_size = SZ_256M > PMD_SIZE ? SZ_256M : PMD_SIZE,
without reformat the patch.
I've sent a v3 with base commit SHA as well.
Please take a look while you're available, thanks!

Best regards,
Richard Cheng.
 
> > 
> > DJ
> > 
> > 
> > 
> > > ---
> > > Changelog:
> > > 
> > > v1 -> v2:
> > >     - Reframe from "fix start alignment" to "enforce PMD invariant for
> > >       volatile CFMWS"
> > >     - Bump cfmws5.window_size to max(SZ_256M, PMD_SIZE)
> > >     - Gate populate_cedt() alignment on
> > >       ACPI_CEDT_CFMWS_RESTRICT_VOLATILE so PMEM windows stay at SZ_256M
> > >     - Add BUILD_BUG_ON for MOCK_AUTO_REGION_SIZE_DEFAULT v.s. PMD_SIZE
> > >       in cxl_test_init()
> > > 
> > > v1 link: https://lore.kernel.org/r/20260519233523.5991-1-icheng@nvidia.com/
> > > 
> > > Best regards,
> > > Richard Cheng.
> > > ---
> > >  tools/testing/cxl/test/cxl.c | 14 ++++++++++++--
> > >  1 file changed, 12 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
> > > index 418669927fb0..09c309af5fd1 100644
> > > --- a/tools/testing/cxl/test/cxl.c
> > > +++ b/tools/testing/cxl/test/cxl.c
> > > @@ -318,7 +318,7 @@ static struct {
> > >  			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_HOSTONLYMEM |
> > >  					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE,
> > >  			.qtg_id = FAKE_QTG_ID,
> > > -			.window_size = SZ_256M,
> > > +			.window_size = SZ_256M > PMD_SIZE ? SZ_256M : PMD_SIZE,
> > >  		},
> > >  		.target = { 3 },
> > >  	},
> > > @@ -495,9 +495,17 @@ static int populate_cedt(void)
> > >  
> > >  	for (i = cfmws_start; i <= cfmws_end; i++) {
> > >  		struct acpi_cedt_cfmws *window = mock_cfmws[i];
> > > +		int align = SZ_256M;
> > >  
> > >  		cfmws_elc_update(window, i);
> > > -		res = alloc_mock_res(window->window_size, SZ_256M);
> > > +		if (window->restrictions & ACPI_CEDT_CFMWS_RESTRICT_VOLATILE)
> > > +			align = max_t(int, SZ_256M, PMD_SIZE);
> > > +		res = alloc_mock_res(window->window_size, align);
> > >  		if (!res)
> > >  			return -ENOMEM;
> > >  		window->base_hpa = res->range.start;
> > > @@ -1826,6 +1834,8 @@ static __init int cxl_test_init(void)
> > >  	int rc, i;
> > >  	struct range mappable;
> > >  
> > > +	BUILD_BUG_ON(!IS_ALIGNED(MOCK_AUTO_REGION_SIZE_DEFAULT, PMD_SIZE));
> > > +
> > >  	cxl_acpi_test();
> > >  	cxl_core_test();
> > >  	cxl_mem_test();
> >

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.