[PATCH 1/2] HMM: test: fix CONFIG_DEVICE_PRIVATE dependencies

[PATCH 1/2] HMM: test: fix CONFIG_DEVICE_PRIVATE dependencies

[Arnd Bergmann]

One cannot select DEVICE_PRIVATE when its dependencies are disabled:

WARNING: unmet direct dependencies detected for DEVICE_PRIVATE
  Depends on [n]: ZONE_DEVICE [=n]
  Selected by [m]:
  - TEST_HMM [=m] && RUNTIME_TESTING_MENU [=y] && TRANSPARENT_HUGEPAGE [=y]
lib/test_hmm.c:480:8: error: implicit declaration of function 'memremap_pages' [-Werror,-Wimplicit-function-declaration]
        ptr = memremap_pages(&devmem->pagemap, numa_node_id());
              ^
lib/test_hmm.c:480:8: note: did you mean 'memcmp_pages'?
include/linux/mm.h:3082:12: note: 'memcmp_pages' declared here
extern int memcmp_pages(struct page *page1, struct page *page2);
           ^
lib/test_hmm.c:480:6: error: incompatible integer to pointer conversion assigning to 'void *' from 'int' [-Werror,-Wint-conversion]
        ptr = memremap_pages(&devmem->pagemap, numa_node_id());
            ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
lib/test_hmm.c:1089:4: error: implicit declaration of function 'memunmap_pages' [-Werror,-Wimplicit-function-declaration]
                        memunmap_pages(&devmem->pagemap);

Add a dependency from CONFIG_TEST_HMM to ZONE_DEVICE, to disallow
those broken configurations.

Fixes: 5d5e54be8a1e ("mm/hmm/test: add selftest driver for HMM")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 lib/Kconfig.debug | 1 +
 1 file changed, 1 insertion(+)

diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index d9885fa37cc0..63667d62c9d8 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -2262,6 +2262,7 @@ config TEST_MEMINIT
 config TEST_HMM
 	tristate "Test HMM (Heterogeneous Memory Management)"
 	depends on TRANSPARENT_HUGEPAGE
+	depends on ZONE_DEVICE
 	select DEVICE_PRIVATE
 	select HMM_MIRROR
 	select MMU_NOTIFIER
-- 
2.26.0

[PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Arnd Bergmann]

CONFIG_DEVICE_PRIVATE cannot be selected in configurations
without ZONE_DEVICE:

WARNING: unmet direct dependencies detected for DEVICE_PRIVATE
  Depends on [n]: ZONE_DEVICE [=n]
  Selected by [y]:
  - DRM_NOUVEAU_SVM [=y] && HAS_IOMEM [=y] && DRM_NOUVEAU [=y] && MMU [=y] && STAGING [=y]
kernel/resource.c:1653:28: error: use of undeclared identifier 'PA_SECTION_SHIFT'
        size = ALIGN(size, 1UL << PA_SECTION_SHIFT);
                                  ^
kernel/resource.c:1654:48: error: use of undeclared identifier 'MAX_PHYSMEM_BITS'

Add a dependency for Nouveau to avoid broken randconfig builds.

Fixes: d2c63df2242e ("mm/hmm: make CONFIG_DEVICE_PRIVATE into a select")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/gpu/drm/nouveau/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/nouveau/Kconfig b/drivers/gpu/drm/nouveau/Kconfig
index af5793f3e7c2..996ec5475908 100644
--- a/drivers/gpu/drm/nouveau/Kconfig
+++ b/drivers/gpu/drm/nouveau/Kconfig
@@ -87,6 +87,7 @@ config DRM_NOUVEAU_BACKLIGHT
 config DRM_NOUVEAU_SVM
 	bool "(EXPERIMENTAL) Enable SVM (Shared Virtual Memory) support"
 	depends on DRM_NOUVEAU
+	depends on ZONE_DEVICE
 	depends on MMU
 	depends on STAGING
 	select DEVICE_PRIVATE
-- 
2.26.0

Re: [PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Jason Gunthorpe]

On Fri, May 08, 2020 at 04:40:09PM +0200, Arnd Bergmann wrote:
> CONFIG_DEVICE_PRIVATE cannot be selected in configurations
> without ZONE_DEVICE:

It is kind of unfortunate to lift dependencies from DEVICE_PRIVATE
into the users, is this really how kconfig is supposed to work or is
something else wrong here?

Jason

Re: [PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Arnd Bergmann]

On Fri, May 8, 2020 at 5:00 PM Jason Gunthorpe <jgg@mellanox.com> wrote:
>
> On Fri, May 08, 2020 at 04:40:09PM +0200, Arnd Bergmann wrote:
> > CONFIG_DEVICE_PRIVATE cannot be selected in configurations
> > without ZONE_DEVICE:
>
> It is kind of unfortunate to lift dependencies from DEVICE_PRIVATE
> into the users, is this really how kconfig is supposed to work or is
> something else wrong here?

Usually the problem is trying to use 'select' for something that can
be done with 'depends on'. I have actually no idea what
CONFIG_DEVICE_PRIVATE does, as it lacks a help text
and is a rather generic term.

Would it be possible to decouple DEVICE_PRIVATE from ZONE_DEVICE?
It sounds like the first is related to the device model, while
the second is for memory management, so maybe the dependency
is not necessary.

       Arnd

Re: [PATCH 1/2] HMM: test: fix CONFIG_DEVICE_PRIVATE dependencies

[Ralph Campbell]

On 5/8/20 7:40 AM, Arnd Bergmann wrote:
> One cannot select DEVICE_PRIVATE when its dependencies are disabled:
> 
> WARNING: unmet direct dependencies detected for DEVICE_PRIVATE
>    Depends on [n]: ZONE_DEVICE [=n]
>    Selected by [m]:
>    - TEST_HMM [=m] && RUNTIME_TESTING_MENU [=y] && TRANSPARENT_HUGEPAGE [=y]
> lib/test_hmm.c:480:8: error: implicit declaration of function 'memremap_pages' [-Werror,-Wimplicit-function-declaration]
>          ptr = memremap_pages(&devmem->pagemap, numa_node_id());
>                ^
> lib/test_hmm.c:480:8: note: did you mean 'memcmp_pages'?
> include/linux/mm.h:3082:12: note: 'memcmp_pages' declared here
> extern int memcmp_pages(struct page *page1, struct page *page2);
>             ^
> lib/test_hmm.c:480:6: error: incompatible integer to pointer conversion assigning to 'void *' from 'int' [-Werror,-Wint-conversion]
>          ptr = memremap_pages(&devmem->pagemap, numa_node_id());
>              ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> lib/test_hmm.c:1089:4: error: implicit declaration of function 'memunmap_pages' [-Werror,-Wimplicit-function-declaration]
>                          memunmap_pages(&devmem->pagemap);
> 
> Add a dependency from CONFIG_TEST_HMM to ZONE_DEVICE, to disallow
> those broken configurations.

Thanks for fixing this.

> 
> Fixes: 5d5e54be8a1e ("mm/hmm/test: add selftest driver for HMM")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Since the HMM tests are not yet in Linus' tree, shouldn't this patch just be folded
into Jason's hmm tree instead of a Fixes: line?

Reviewed-by: Ralph Campbell <rcampbell@nvidia.com>

> ---
>   lib/Kconfig.debug | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
> index d9885fa37cc0..63667d62c9d8 100644
> --- a/lib/Kconfig.debug
> +++ b/lib/Kconfig.debug
> @@ -2262,6 +2262,7 @@ config TEST_MEMINIT
>   config TEST_HMM
>   	tristate "Test HMM (Heterogeneous Memory Management)"
>   	depends on TRANSPARENT_HUGEPAGE
> +	depends on ZONE_DEVICE
>   	select DEVICE_PRIVATE
>   	select HMM_MIRROR
>   	select MMU_NOTIFIER
> 

Re: [PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Randy Dunlap]

On 5/8/20 7:40 AM, Arnd Bergmann wrote:
> CONFIG_DEVICE_PRIVATE cannot be selected in configurations
> without ZONE_DEVICE:
> 
> WARNING: unmet direct dependencies detected for DEVICE_PRIVATE
>   Depends on [n]: ZONE_DEVICE [=n]
>   Selected by [y]:
>   - DRM_NOUVEAU_SVM [=y] && HAS_IOMEM [=y] && DRM_NOUVEAU [=y] && MMU [=y] && STAGING [=y]
> kernel/resource.c:1653:28: error: use of undeclared identifier 'PA_SECTION_SHIFT'
>         size = ALIGN(size, 1UL << PA_SECTION_SHIFT);
>                                   ^
> kernel/resource.c:1654:48: error: use of undeclared identifier 'MAX_PHYSMEM_BITS'
> 
> Add a dependency for Nouveau to avoid broken randconfig builds.
> 
> Fixes: d2c63df2242e ("mm/hmm: make CONFIG_DEVICE_PRIVATE into a select")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested

Thanks.

> ---
>  drivers/gpu/drm/nouveau/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/gpu/drm/nouveau/Kconfig b/drivers/gpu/drm/nouveau/Kconfig
> index af5793f3e7c2..996ec5475908 100644
> --- a/drivers/gpu/drm/nouveau/Kconfig
> +++ b/drivers/gpu/drm/nouveau/Kconfig
> @@ -87,6 +87,7 @@ config DRM_NOUVEAU_BACKLIGHT
>  config DRM_NOUVEAU_SVM
>  	bool "(EXPERIMENTAL) Enable SVM (Shared Virtual Memory) support"
>  	depends on DRM_NOUVEAU
> +	depends on ZONE_DEVICE
>  	depends on MMU
>  	depends on STAGING
>  	select DEVICE_PRIVATE
> 


-- 
~Randy

Re: [PATCH 1/2] HMM: test: fix CONFIG_DEVICE_PRIVATE dependencies

[Randy Dunlap]

On 5/8/20 7:40 AM, Arnd Bergmann wrote:
> One cannot select DEVICE_PRIVATE when its dependencies are disabled:
> 
> WARNING: unmet direct dependencies detected for DEVICE_PRIVATE
>   Depends on [n]: ZONE_DEVICE [=n]
>   Selected by [m]:
>   - TEST_HMM [=m] && RUNTIME_TESTING_MENU [=y] && TRANSPARENT_HUGEPAGE [=y]
> lib/test_hmm.c:480:8: error: implicit declaration of function 'memremap_pages' [-Werror,-Wimplicit-function-declaration]
>         ptr = memremap_pages(&devmem->pagemap, numa_node_id());
>               ^
> lib/test_hmm.c:480:8: note: did you mean 'memcmp_pages'?
> include/linux/mm.h:3082:12: note: 'memcmp_pages' declared here
> extern int memcmp_pages(struct page *page1, struct page *page2);
>            ^
> lib/test_hmm.c:480:6: error: incompatible integer to pointer conversion assigning to 'void *' from 'int' [-Werror,-Wint-conversion]
>         ptr = memremap_pages(&devmem->pagemap, numa_node_id());
>             ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> lib/test_hmm.c:1089:4: error: implicit declaration of function 'memunmap_pages' [-Werror,-Wimplicit-function-declaration]
>                         memunmap_pages(&devmem->pagemap);
> 
> Add a dependency from CONFIG_TEST_HMM to ZONE_DEVICE, to disallow
> those broken configurations.
> 
> Fixes: 5d5e54be8a1e ("mm/hmm/test: add selftest driver for HMM")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested

Thanks.

> ---
>  lib/Kconfig.debug | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
> index d9885fa37cc0..63667d62c9d8 100644
> --- a/lib/Kconfig.debug
> +++ b/lib/Kconfig.debug
> @@ -2262,6 +2262,7 @@ config TEST_MEMINIT
>  config TEST_HMM
>  	tristate "Test HMM (Heterogeneous Memory Management)"
>  	depends on TRANSPARENT_HUGEPAGE
> +	depends on ZONE_DEVICE
>  	select DEVICE_PRIVATE
>  	select HMM_MIRROR
>  	select MMU_NOTIFIER
> 


-- 
~Randy

Re: [PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Jason Gunthorpe]

On Fri, May 08, 2020 at 05:05:03PM +0200, Arnd Bergmann wrote:
> On Fri, May 8, 2020 at 5:00 PM Jason Gunthorpe <jgg@mellanox.com> wrote:
> >
> > On Fri, May 08, 2020 at 04:40:09PM +0200, Arnd Bergmann wrote:
> > > CONFIG_DEVICE_PRIVATE cannot be selected in configurations
> > > without ZONE_DEVICE:
> >
> > It is kind of unfortunate to lift dependencies from DEVICE_PRIVATE
> > into the users, is this really how kconfig is supposed to work or is
> > something else wrong here?
> 
> Usually the problem is trying to use 'select' for something that can
> be done with 'depends on'. I have actually no idea what
> CONFIG_DEVICE_PRIVATE does, as it lacks a help text
> and is a rather generic term.

It is a sub mode of ZONE_DEVICE, ie DEVICE_PRIVATE turns on a certain
kind of ZONE_DEVICE page.

Both ZONE_DEVICE and DEVICE_PRIVATE are APIs families drivers use,
there is no reason for a user to select either of these directly.

Jason

Re: [PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Arnd Bergmann]

On Fri, May 8, 2020 at 9:04 PM Jason Gunthorpe <jgg@mellanox.com> wrote:
>
> On Fri, May 08, 2020 at 05:05:03PM +0200, Arnd Bergmann wrote:
> > On Fri, May 8, 2020 at 5:00 PM Jason Gunthorpe <jgg@mellanox.com> wrote:
> > >
> > > On Fri, May 08, 2020 at 04:40:09PM +0200, Arnd Bergmann wrote:
> > > > CONFIG_DEVICE_PRIVATE cannot be selected in configurations
> > > > without ZONE_DEVICE:
> > >
> > > It is kind of unfortunate to lift dependencies from DEVICE_PRIVATE
> > > into the users, is this really how kconfig is supposed to work or is
> > > something else wrong here?
> >
> > Usually the problem is trying to use 'select' for something that can
> > be done with 'depends on'. I have actually no idea what
> > CONFIG_DEVICE_PRIVATE does, as it lacks a help text
> > and is a rather generic term.
>
> It is a sub mode of ZONE_DEVICE, ie DEVICE_PRIVATE turns on a certain
> kind of ZONE_DEVICE page.
>
> Both ZONE_DEVICE and DEVICE_PRIVATE are APIs families drivers use,
> there is no reason for a user to select either of these directly.

Ok, then how about making ZONE_DEVICE a hidden symbol and adding
something like

config ZONE_DEVICE_POSSIBLE
        def_bool y
        depends on MEMORY_HOTPLUG
        depends on MEMORY_HOTREMOVE
        depends on SPARSEMEM_VMEMMAP
        depends on ARCH_HAS_PTE_DEVMAP

then drivers that want it can do

config FOO
        tristate "user visible option"
        depends on ZONE_DEVICE_POSSIBLE
        select ZONE_DEVICE

      Arnd

Re: [PATCH 2/2] nouveau: fix dependencies for DEVICE_PRIVATE

[Jason Gunthorpe]

On Fri, May 08, 2020 at 04:40:09PM +0200, Arnd Bergmann wrote:
> CONFIG_DEVICE_PRIVATE cannot be selected in configurations
> without ZONE_DEVICE:
> 
> WARNING: unmet direct dependencies detected for DEVICE_PRIVATE
>   Depends on [n]: ZONE_DEVICE [=n]
>   Selected by [y]:
>   - DRM_NOUVEAU_SVM [=y] && HAS_IOMEM [=y] && DRM_NOUVEAU [=y] && MMU [=y] && STAGING [=y]
> kernel/resource.c:1653:28: error: use of undeclared identifier 'PA_SECTION_SHIFT'
>         size = ALIGN(size, 1UL << PA_SECTION_SHIFT);
>                                   ^
> kernel/resource.c:1654:48: error: use of undeclared identifier 'MAX_PHYSMEM_BITS'
> 
> Add a dependency for Nouveau to avoid broken randconfig builds.
> 
> Fixes: d2c63df2242e ("mm/hmm: make CONFIG_DEVICE_PRIVATE into a
> select")

I've reverted the patch this fixes, it seems more trouble than it is
worth.

Thanks,
Jason