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 <[email protected]>
---
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/[email protected]
- 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
On Wed, 22 May 2024 16:38:01 -0400 Eric Chanudet <[email protected]> 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.
Eric Chanudet <[email protected]> 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 <[email protected]>
>
> ---
> 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/[email protected]
> - 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 <[email protected]> (powerpc)
cheers
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 <[email protected]>
>
> ---
> 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/[email protected]
> - 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.