[PATCH v2] mm/mm_init: use node's number of cpus in deferred_page_init_max_threads

[PATCH v2] mm/mm_init: use node's number of cpus in deferred_page_init_max_threads

[Eric Chanudet]

x86_64 is already using the node's cpu as maximum threads. Make that the
default for all archs setting DEFERRED_STRUCT_PAGE_INIT.

This returns to the behavior prior making the function arch-specific
with commit ecd096506922 ("mm: make deferred init's max threads
arch-specific").

Signed-off-by: Eric Chanudet <echanude@redhat.com>

---
Setting DEFERRED_STRUCT_PAGE_INIT and testing on a few arm64 platforms
shows faster deferred_init_memmap completions:

|         | x13s        | SA8775p-ride | Ampere R137-P31 | Ampere HR330 |
|         | Metal, 32GB | VM, 36GB     | VM, 58GB        | Metal, 128GB |
|         | 8cpus       | 8cpus        | 8cpus           | 32cpus       |
|---------|-------------|--------------|-----------------|--------------|
| threads |  ms     (%) | ms       (%) |  ms         (%) |  ms      (%) |
|---------|-------------|--------------|-----------------|--------------|
| 1       | 108    (0%) | 72      (0%) | 224        (0%) | 324     (0%) |
| cpus    |  24  (-77%) | 36    (-50%) |  40      (-82%) |  56   (-82%) |

- v1: https://lore.kernel.org/linux-arm-kernel/20240520231555.395979-5-echanude@redhat.com
- Changes since v1:
 - Make the generic function return the number of cpus of the node as
   max threads limit instead overriding it for arm64.
- Drop Baoquan He's R-b on v1 since the logic changed.
- Add CCs according to patch changes (ppc and s390 set
  DEFERRED_STRUCT_PAGE_INIT by default).

 arch/x86/mm/init_64.c | 12 ------------
 mm/mm_init.c          |  2 +-
 2 files changed, 1 insertion(+), 13 deletions(-)

diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 7e177856ee4f..adec42928ec1 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -1354,18 +1354,6 @@ void __init mem_init(void)
 	preallocate_vmalloc_pages();
 }
 
-#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
-int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask)
-{
-	/*
-	 * More CPUs always led to greater speedups on tested systems, up to
-	 * all the nodes' CPUs.  Use all since the system is otherwise idle
-	 * now.
-	 */
-	return max_t(int, cpumask_weight(node_cpumask), 1);
-}
-#endif
-
 int kernel_set_to_readonly;
 
 void mark_rodata_ro(void)
diff --git a/mm/mm_init.c b/mm/mm_init.c
index f72b852bd5b8..e0023aa68555 100644
--- a/mm/mm_init.c
+++ b/mm/mm_init.c
@@ -2126,7 +2126,7 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
 __weak int __init
 deferred_page_init_max_threads(const struct cpumask *node_cpumask)
 {
-	return 1;
+	return max_t(int, cpumask_weight(node_cpumask), 1);
 }
 
 /* Initialise remaining memory on a node */
-- 
2.44.0

Re: [PATCH v2] mm/mm_init: use node's number of cpus in deferred_page_init_max_threads

[Andrew Morton]

On Wed, 22 May 2024 16:38:01 -0400 Eric Chanudet <echanude@redhat.com> wrote:

> x86_64 is already using the node's cpu as maximum threads. Make that the
> default for all archs setting DEFERRED_STRUCT_PAGE_INIT.
> 
> This returns to the behavior prior making the function arch-specific
> with commit ecd096506922 ("mm: make deferred init's max threads
> arch-specific").
> 

It isn't clear to me what is the runtime effect of this change upon our
users.  Can you please prepare a sentence which spells this out?

> 
> ---
> Setting DEFERRED_STRUCT_PAGE_INIT and testing on a few arm64 platforms
> shows faster deferred_init_memmap completions:
> 
> |         | x13s        | SA8775p-ride | Ampere R137-P31 | Ampere HR330 |
> |         | Metal, 32GB | VM, 36GB     | VM, 58GB        | Metal, 128GB |
> |         | 8cpus       | 8cpus        | 8cpus           | 32cpus       |
> |---------|-------------|--------------|-----------------|--------------|
> | threads |  ms     (%) | ms       (%) |  ms         (%) |  ms      (%) |
> |---------|-------------|--------------|-----------------|--------------|
> | 1       | 108    (0%) | 72      (0%) | 224        (0%) | 324     (0%) |
> | cpus    |  24  (-77%) | 36    (-50%) |  40      (-82%) |  56   (-82%) |

The above is useful info, I'll hoist it into the main changelog.

> --- a/mm/mm_init.c
> +++ b/mm/mm_init.c
> @@ -2126,7 +2126,7 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
>  __weak int __init
>  deferred_page_init_max_threads(const struct cpumask *node_cpumask)
>  {
> -	return 1;
> +	return max_t(int, cpumask_weight(node_cpumask), 1);
>  }

It's an unrelated cleanup , but that could be

	max(cpumask_weight(node_cpumask), 1U);

and the function could/should return unsigned.

Re: [PATCH v2] mm/mm_init: use node's number of cpus in deferred_page_init_max_threads

[Michael Ellerman]

Eric Chanudet <echanude@redhat.com> writes:
> x86_64 is already using the node's cpu as maximum threads. Make that the
> default for all archs setting DEFERRED_STRUCT_PAGE_INIT.
>
> This returns to the behavior prior making the function arch-specific
> with commit ecd096506922 ("mm: make deferred init's max threads
> arch-specific").
>
> Signed-off-by: Eric Chanudet <echanude@redhat.com>
>
> ---
> Setting DEFERRED_STRUCT_PAGE_INIT and testing on a few arm64 platforms
> shows faster deferred_init_memmap completions:
>
> |         | x13s        | SA8775p-ride | Ampere R137-P31 | Ampere HR330 |
> |         | Metal, 32GB | VM, 36GB     | VM, 58GB        | Metal, 128GB |
> |         | 8cpus       | 8cpus        | 8cpus           | 32cpus       |
> |---------|-------------|--------------|-----------------|--------------|
> | threads |  ms     (%) | ms       (%) |  ms         (%) |  ms      (%) |
> |---------|-------------|--------------|-----------------|--------------|
> | 1       | 108    (0%) | 72      (0%) | 224        (0%) | 324     (0%) |
> | cpus    |  24  (-77%) | 36    (-50%) |  40      (-82%) |  56   (-82%) |
>
> - v1: https://lore.kernel.org/linux-arm-kernel/20240520231555.395979-5-echanude@redhat.com
> - Changes since v1:
>  - Make the generic function return the number of cpus of the node as
>    max threads limit instead overriding it for arm64.
> - Drop Baoquan He's R-b on v1 since the logic changed.
> - Add CCs according to patch changes (ppc and s390 set
>   DEFERRED_STRUCT_PAGE_INIT by default).
>
>  arch/x86/mm/init_64.c | 12 ------------
>  mm/mm_init.c          |  2 +-
>  2 files changed, 1 insertion(+), 13 deletions(-)

On a machine here (1TB, 40 cores, 4KB pages) the existing code gives:

  [    0.500124] node 2 deferred pages initialised in 210ms
  [    0.515790] node 3 deferred pages initialised in 230ms
  [    0.516061] node 0 deferred pages initialised in 230ms
  [    0.516522] node 7 deferred pages initialised in 230ms
  [    0.516672] node 4 deferred pages initialised in 230ms
  [    0.516798] node 6 deferred pages initialised in 230ms
  [    0.517051] node 5 deferred pages initialised in 230ms
  [    0.523887] node 1 deferred pages initialised in 240ms

vs with the patch:

  [    0.379613] node 0 deferred pages initialised in 90ms
  [    0.380388] node 1 deferred pages initialised in 90ms
  [    0.380540] node 4 deferred pages initialised in 100ms
  [    0.390239] node 6 deferred pages initialised in 100ms
  [    0.390249] node 2 deferred pages initialised in 100ms
  [    0.390786] node 3 deferred pages initialised in 110ms
  [    0.396721] node 5 deferred pages initialised in 110ms
  [    0.397095] node 7 deferred pages initialised in 110ms

Which is a nice speedup.

Tested-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)

cheers

Re: [PATCH v2] mm/mm_init: use node's number of cpus in deferred_page_init_max_threads

[Mike Rapoport]

On Wed, May 22, 2024 at 04:38:01PM -0400, Eric Chanudet wrote:
> x86_64 is already using the node's cpu as maximum threads. Make that the
> default for all archs setting DEFERRED_STRUCT_PAGE_INIT.
> 
> This returns to the behavior prior making the function arch-specific
> with commit ecd096506922 ("mm: make deferred init's max threads
> arch-specific").
> 
> Signed-off-by: Eric Chanudet <echanude@redhat.com>
> 
> ---
> Setting DEFERRED_STRUCT_PAGE_INIT and testing on a few arm64 platforms
> shows faster deferred_init_memmap completions:
> 
> |         | x13s        | SA8775p-ride | Ampere R137-P31 | Ampere HR330 |
> |         | Metal, 32GB | VM, 36GB     | VM, 58GB        | Metal, 128GB |
> |         | 8cpus       | 8cpus        | 8cpus           | 32cpus       |
> |---------|-------------|--------------|-----------------|--------------|
> | threads |  ms     (%) | ms       (%) |  ms         (%) |  ms      (%) |
> |---------|-------------|--------------|-----------------|--------------|
> | 1       | 108    (0%) | 72      (0%) | 224        (0%) | 324     (0%) |
> | cpus    |  24  (-77%) | 36    (-50%) |  40      (-82%) |  56   (-82%) |
> 
> - v1: https://lore.kernel.org/linux-arm-kernel/20240520231555.395979-5-echanude@redhat.com
> - Changes since v1:
>  - Make the generic function return the number of cpus of the node as
>    max threads limit instead overriding it for arm64.
> - Drop Baoquan He's R-b on v1 since the logic changed.
> - Add CCs according to patch changes (ppc and s390 set
>   DEFERRED_STRUCT_PAGE_INIT by default).
> 
>  arch/x86/mm/init_64.c | 12 ------------
>  mm/mm_init.c          |  2 +-
>  2 files changed, 1 insertion(+), 13 deletions(-)
> 
> diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
> index 7e177856ee4f..adec42928ec1 100644
> --- a/arch/x86/mm/init_64.c
> +++ b/arch/x86/mm/init_64.c
> @@ -1354,18 +1354,6 @@ void __init mem_init(void)
>  	preallocate_vmalloc_pages();
>  }
>  
> -#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
> -int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask)
> -{
> -	/*
> -	 * More CPUs always led to greater speedups on tested systems, up to
> -	 * all the nodes' CPUs.  Use all since the system is otherwise idle
> -	 * now.
> -	 */
> -	return max_t(int, cpumask_weight(node_cpumask), 1);
> -}
> -#endif
> -
>  int kernel_set_to_readonly;
>  
>  void mark_rodata_ro(void)
> diff --git a/mm/mm_init.c b/mm/mm_init.c
> index f72b852bd5b8..e0023aa68555 100644
> --- a/mm/mm_init.c
> +++ b/mm/mm_init.c
> @@ -2126,7 +2126,7 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
>  __weak int __init

If s390 folks confirm there's no regression for them I think we can make
this static.

>  deferred_page_init_max_threads(const struct cpumask *node_cpumask)
>  {
> -	return 1;
> +	return max_t(int, cpumask_weight(node_cpumask), 1);
>  }
>  
>  /* Initialise remaining memory on a node */
> -- 
> 2.44.0
> 

-- 
Sincerely yours,
Mike.