[PATCH v5 00/12] mm: establish const-correctness for pointer parameters

Max Kellermann posted 12 patches 3 months, 1 week ago
There is a newer version of this series
arch/arm/include/asm/highmem.h      |  6 +--
arch/parisc/include/asm/processor.h |  2 +-
arch/parisc/kernel/sys_parisc.c     |  2 +-
arch/s390/mm/mmap.c                 |  7 ++--
arch/sparc/kernel/sys_sparc_64.c    |  3 +-
arch/x86/mm/mmap.c                  |  7 ++--
arch/xtensa/include/asm/highmem.h   |  2 +-
include/linux/fs.h                  |  7 ++--
include/linux/highmem-internal.h    | 44 +++++++++++----------
include/linux/highmem.h             |  8 ++--
include/linux/mm.h                  | 56 +++++++++++++--------------
include/linux/mm_inline.h           | 26 +++++++------
include/linux/mm_types.h            |  4 +-
include/linux/mmzone.h              | 42 ++++++++++----------
include/linux/pagemap.h             | 59 +++++++++++++++--------------
include/linux/sched/mm.h            |  4 +-
include/linux/shmem_fs.h            |  4 +-
mm/highmem.c                        | 10 ++---
mm/oom_kill.c                       |  7 ++--
mm/shmem.c                          |  6 +--
mm/util.c                           | 20 ++++++----
21 files changed, 171 insertions(+), 155 deletions(-)
[PATCH v5 00/12] mm: establish const-correctness for pointer parameters
Posted by Max Kellermann 3 months, 1 week ago
For improved const-correctness.

This patch series systematically adds const qualifiers to pointer
parameters throughout the memory management subsystem, establishing a
foundation for improved const-correctness across the entire Linux
kernel.

Const-correctness provides multiple benefits:

1. Type Safety: The compiler enforces that functions marked as taking
   const parameters cannot accidentally modify the data, catching
   potential bugs at compile time rather than runtime.

2. Compiler Optimizations: When the compiler knows data won't be
   modified, it can generate more efficient code through better
   register allocation, code motion, and aliasing analysis.

3. API Documentation: Const qualifiers serve as self-documenting code,
   making it immediately clear to developers which functions are
   read-only operations versus those that modify state.

4. Maintenance Safety: Future modifications to const-correct code are
   less likely to introduce subtle bugs, as the compiler will reject
   attempts to modify data that should remain unchanged.

The memory management subsystem is a fundamental building block of the
kernel.  Most higher-level kernel subsystems (filesystems, drivers,
networking) depend on mm interfaces.  By establishing
const-correctness at this foundational level:

1. Enables Propagation: Higher-level subsystems can adopt
   const-correctness in their own interfaces.  Without const-correct
   mm functions, filesystems cannot mark their own parameters const
   when they need to call mm functions.

2. Maximum Impact: Changes to core mm APIs benefit the entire kernel, as
   these functions are called from virtually every subsystem.

3. Prevents Impedance Mismatch: Without const-correctness in mm, other
   subsystems must either cast away const (dangerous) or avoid using
   const altogether (missing optimization opportunities).

Each patch focuses on a specific header or subsystem component to ease review
and bisection.

This work was initially posted as a single large patch:
 https://lore.kernel.org/lkml/20250827192233.447920-1-max.kellermann@ionos.com/

Following feedback from Lorenzo Stoakes and David Hildenbrand, it has been
split into focused, reviewable chunks. The approach was validated with a
smaller patch that received agreement:
 https://lore.kernel.org/lkml/20250828130311.772993-1-max.kellermann@ionos.com/

Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
---
v1 -> v2:
- made several parameter values const (i.e. the pointer address, not
  just the pointed-to memory), as suggested by Andrew Morton and
  Yuanchu Xie
- drop existing+obsolete "extern" keywords on lines modified by these
  patches (suggested by Vishal Moola)
- add missing parameter names on lines modified by these patches
  (suggested by Vishal Moola)
- more "const" pointers (e.g. the task_struct passed to
  process_shares_mm())
- add missing "const" to s390, fixing s390 build failure
- moved the mmap_is_legacy() change in arch/s390/mm/mmap.c from 08/12
  to 06/12 (suggested by Vishal Moola)

v2 -> v3:
- remove garbage from 06/12
- changed tags on subject line (suggested by Matthew Wilcox)

v3 -> v4:
- more verbose commit messages including a listing of function names
  (suggested by David Hildenbrand and Lorenzo Stoakes)

v4 -> v5:
- back to shorter commit messages after an agreement between David
  Hildenbrand and Lorenzo Stoakes was found

Max Kellermann (12):
  mm: constify shmem related test functions for improved
    const-correctness
  mm: constify pagemap related test functions for improved
    const-correctness
  mm: constify zone related test functions for improved
    const-correctness
  fs: constify mapping related test functions for improved
    const-correctness
  mm: constify process_shares_mm() for improved const-correctness
  mm, s390: constify mapping related test functions for improved
    const-correctness
  parisc: constify mmap_upper_limit() parameter for improved
    const-correctness
  mm: constify arch_pick_mmap_layout() for improved const-correctness
  mm: constify ptdesc_pmd_pts_count() and folio_get_private()
  mm: constify various inline test functions for improved
    const-correctness
  mm: constify assert/test functions in mm.h
  mm: constify highmem related functions for improved const-correctness

 arch/arm/include/asm/highmem.h      |  6 +--
 arch/parisc/include/asm/processor.h |  2 +-
 arch/parisc/kernel/sys_parisc.c     |  2 +-
 arch/s390/mm/mmap.c                 |  7 ++--
 arch/sparc/kernel/sys_sparc_64.c    |  3 +-
 arch/x86/mm/mmap.c                  |  7 ++--
 arch/xtensa/include/asm/highmem.h   |  2 +-
 include/linux/fs.h                  |  7 ++--
 include/linux/highmem-internal.h    | 44 +++++++++++----------
 include/linux/highmem.h             |  8 ++--
 include/linux/mm.h                  | 56 +++++++++++++--------------
 include/linux/mm_inline.h           | 26 +++++++------
 include/linux/mm_types.h            |  4 +-
 include/linux/mmzone.h              | 42 ++++++++++----------
 include/linux/pagemap.h             | 59 +++++++++++++++--------------
 include/linux/sched/mm.h            |  4 +-
 include/linux/shmem_fs.h            |  4 +-
 mm/highmem.c                        | 10 ++---
 mm/oom_kill.c                       |  7 ++--
 mm/shmem.c                          |  6 +--
 mm/util.c                           | 20 ++++++----
 21 files changed, 171 insertions(+), 155 deletions(-)

-- 
2.47.2
Re: [PATCH v5 00/12] mm: establish const-correctness for pointer parameters
Posted by Lorenzo Stoakes 3 months, 1 week ago
On Mon, Sep 01, 2025 at 02:30:16PM +0200, Max Kellermann wrote:
> For improved const-correctness.
>
> This patch series systematically adds const qualifiers to pointer
> parameters throughout the memory management subsystem, establishing a
> foundation for improved const-correctness across the entire Linux
> kernel.
>
> Const-correctness provides multiple benefits:
>
> 1. Type Safety: The compiler enforces that functions marked as taking
>    const parameters cannot accidentally modify the data, catching
>    potential bugs at compile time rather than runtime.
>
> 2. Compiler Optimizations: When the compiler knows data won't be
>    modified, it can generate more efficient code through better
>    register allocation, code motion, and aliasing analysis.
>
> 3. API Documentation: Const qualifiers serve as self-documenting code,
>    making it immediately clear to developers which functions are
>    read-only operations versus those that modify state.
>
> 4. Maintenance Safety: Future modifications to const-correct code are
>    less likely to introduce subtle bugs, as the compiler will reject
>    attempts to modify data that should remain unchanged.

I think all of the above is really a lot of noise that doesn't add a huge
amount of value.

Please in your own words say why you are doing it, and also please mention
why you feel it's justified to do:

const <type> *const param

As mentioned on review of 2/12.

I'm pretty well leaning towards - let's just not do the 2nd const at all,
unless there's a really good reason to do so.

There are also legit cases where you might want to reassign a local
variable.

Largely, granted, you shouldn't be reassinging params, but it's pretty
constraining.

>
> The memory management subsystem is a fundamental building block of the
> kernel.  Most higher-level kernel subsystems (filesystems, drivers,
> networking) depend on mm interfaces.  By establishing
> const-correctness at this foundational level:
>
> 1. Enables Propagation: Higher-level subsystems can adopt
>    const-correctness in their own interfaces.  Without const-correct
>    mm functions, filesystems cannot mark their own parameters const
>    when they need to call mm functions.
>
> 2. Maximum Impact: Changes to core mm APIs benefit the entire kernel, as
>    these functions are called from virtually every subsystem.
>
> 3. Prevents Impedance Mismatch: Without const-correctness in mm, other
>    subsystems must either cast away const (dangerous) or avoid using
>    const altogether (missing optimization opportunities).
>
> Each patch focuses on a specific header or subsystem component to ease review
> and bisection.

All this is unnecessary noise, can you summarise more succinctly.

More words than 'const-ify everything' doesn't mean 'several paragraphs of
noise'.

>
> This work was initially posted as a single large patch:
>  https://lore.kernel.org/lkml/20250827192233.447920-1-max.kellermann@ionos.com/
>
> Following feedback from Lorenzo Stoakes and David Hildenbrand, it has been
> split into focused, reviewable chunks. The approach was validated with a
> smaller patch that received agreement:
>  https://lore.kernel.org/lkml/20250828130311.772993-1-max.kellermann@ionos.com/
>

^--- All of this should be below the line and is associated with versions
not the series as a whole really.

> Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
> ---
> v1 -> v2:
> - made several parameter values const (i.e. the pointer address, not
>   just the pointed-to memory), as suggested by Andrew Morton and
>   Yuanchu Xie
> - drop existing+obsolete "extern" keywords on lines modified by these
>   patches (suggested by Vishal Moola)
> - add missing parameter names on lines modified by these patches
>   (suggested by Vishal Moola)
> - more "const" pointers (e.g. the task_struct passed to
>   process_shares_mm())
> - add missing "const" to s390, fixing s390 build failure
> - moved the mmap_is_legacy() change in arch/s390/mm/mmap.c from 08/12
>   to 06/12 (suggested by Vishal Moola)
>
> v2 -> v3:
> - remove garbage from 06/12
> - changed tags on subject line (suggested by Matthew Wilcox)
>
> v3 -> v4:
> - more verbose commit messages including a listing of function names
>   (suggested by David Hildenbrand and Lorenzo Stoakes)
>
> v4 -> v5:
> - back to shorter commit messages after an agreement between David
>   Hildenbrand and Lorenzo Stoakes was found

I did ask you to do this in reverse order and add lore links :) I mean
these aren't big things but it's REALLY helpful for reviewers.

Thanks.

>
> Max Kellermann (12):
>   mm: constify shmem related test functions for improved
>     const-correctness
>   mm: constify pagemap related test functions for improved
>     const-correctness
>   mm: constify zone related test functions for improved
>     const-correctness
>   fs: constify mapping related test functions for improved
>     const-correctness
>   mm: constify process_shares_mm() for improved const-correctness
>   mm, s390: constify mapping related test functions for improved
>     const-correctness
>   parisc: constify mmap_upper_limit() parameter for improved
>     const-correctness
>   mm: constify arch_pick_mmap_layout() for improved const-correctness
>   mm: constify ptdesc_pmd_pts_count() and folio_get_private()
>   mm: constify various inline test functions for improved
>     const-correctness
>   mm: constify assert/test functions in mm.h
>   mm: constify highmem related functions for improved const-correctness
>
>  arch/arm/include/asm/highmem.h      |  6 +--
>  arch/parisc/include/asm/processor.h |  2 +-
>  arch/parisc/kernel/sys_parisc.c     |  2 +-
>  arch/s390/mm/mmap.c                 |  7 ++--
>  arch/sparc/kernel/sys_sparc_64.c    |  3 +-
>  arch/x86/mm/mmap.c                  |  7 ++--
>  arch/xtensa/include/asm/highmem.h   |  2 +-
>  include/linux/fs.h                  |  7 ++--
>  include/linux/highmem-internal.h    | 44 +++++++++++----------
>  include/linux/highmem.h             |  8 ++--
>  include/linux/mm.h                  | 56 +++++++++++++--------------
>  include/linux/mm_inline.h           | 26 +++++++------
>  include/linux/mm_types.h            |  4 +-
>  include/linux/mmzone.h              | 42 ++++++++++----------
>  include/linux/pagemap.h             | 59 +++++++++++++++--------------
>  include/linux/sched/mm.h            |  4 +-
>  include/linux/shmem_fs.h            |  4 +-
>  mm/highmem.c                        | 10 ++---
>  mm/oom_kill.c                       |  7 ++--
>  mm/shmem.c                          |  6 +--
>  mm/util.c                           | 20 ++++++----
>  21 files changed, 171 insertions(+), 155 deletions(-)
>
> --
> 2.47.2
>
Re: [PATCH v5 00/12] mm: establish const-correctness for pointer parameters
Posted by Matthew Wilcox 3 months, 1 week ago
On Mon, Sep 01, 2025 at 02:30:16PM +0200, Max Kellermann wrote:
> For improved const-correctness.

SLOW DOWN.

This series is unimportant churn.  There's no way it should be up to v5
already.  Wait a freaking week before you send another version.
Re: [PATCH v5 00/12] mm: establish const-correctness for pointer parameters
Posted by David Hildenbrand 3 months, 1 week ago
On 02.09.25 04:50, Matthew Wilcox wrote:
> On Mon, Sep 01, 2025 at 02:30:16PM +0200, Max Kellermann wrote:
>> For improved const-correctness.
> 
> SLOW DOWN.
> 
> This series is unimportant churn.  There's no way it should be up to v5
> already.  Wait a freaking week before you send another version.

... while we have v6 already on the list :D

In this case, it was reasonable to follow up rather sooner than later 
(too much drama and v6 now seems to make people happy, I wouldn't want 
that spanning over multiple weeks).

But yeah, too many revisions on a single day.

-- 
Cheers

David / dhildenb
Re: [PATCH v5 00/12] mm: establish const-correctness for pointer parameters
Posted by Christian Brauner 3 months, 1 week ago
On Mon, Sep 01, 2025 at 02:30:16PM +0200, Max Kellermann wrote:
> For improved const-correctness.
> 
> This patch series systematically adds const qualifiers to pointer
> parameters throughout the memory management subsystem, establishing a
> foundation for improved const-correctness across the entire Linux
> kernel.
> 
> Const-correctness provides multiple benefits:
> 
> 1. Type Safety: The compiler enforces that functions marked as taking
>    const parameters cannot accidentally modify the data, catching
>    potential bugs at compile time rather than runtime.
> 
> 2. Compiler Optimizations: When the compiler knows data won't be
>    modified, it can generate more efficient code through better
>    register allocation, code motion, and aliasing analysis.
> 
> 3. API Documentation: Const qualifiers serve as self-documenting code,
>    making it immediately clear to developers which functions are
>    read-only operations versus those that modify state.
> 
> 4. Maintenance Safety: Future modifications to const-correct code are
>    less likely to introduce subtle bugs, as the compiler will reject
>    attempts to modify data that should remain unchanged.
> 
> The memory management subsystem is a fundamental building block of the
> kernel.  Most higher-level kernel subsystems (filesystems, drivers,
> networking) depend on mm interfaces.  By establishing
> const-correctness at this foundational level:
> 
> 1. Enables Propagation: Higher-level subsystems can adopt
>    const-correctness in their own interfaces.  Without const-correct
>    mm functions, filesystems cannot mark their own parameters const
>    when they need to call mm functions.
> 
> 2. Maximum Impact: Changes to core mm APIs benefit the entire kernel, as
>    these functions are called from virtually every subsystem.
> 
> 3. Prevents Impedance Mismatch: Without const-correctness in mm, other
>    subsystems must either cast away const (dangerous) or avoid using
>    const altogether (missing optimization opportunities).
> 
> Each patch focuses on a specific header or subsystem component to ease review
> and bisection.
> 
> This work was initially posted as a single large patch:
>  https://lore.kernel.org/lkml/20250827192233.447920-1-max.kellermann@ionos.com/
> 
> Following feedback from Lorenzo Stoakes and David Hildenbrand, it has been
> split into focused, reviewable chunks. The approach was validated with a
> smaller patch that received agreement:
>  https://lore.kernel.org/lkml/20250828130311.772993-1-max.kellermann@ionos.com/
> 
> Signed-off-by: Max Kellermann <max.kellermann@ionos.com>

Seems fine,
Reviewed-by: Christian Brauner <brauner@kernel.org>