| Message ID | 20240528185455.643227-4-echanude@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v3] mm/mm_init: use node's number of cpus in deferred_page_init_max_threads | expand |
On Tue, May 28, 2024 at 02:54:58PM -0400, Eric Chanudet wrote: > When DEFERRED_STRUCT_PAGE_INIT=y, use a node's cpu count as maximum > thread count for the deferred initialization of struct pages via padata. > This should result in shorter boot times for these configurations by > going through page_alloc_init_late() faster as systems tend not to be > under heavy load that early in the bootstrap. > > Only x86_64 does that now. Make it archs agnostic when > DEFERRED_STRUCT_PAGE_INIT is set. With the default defconfigs, that > includes powerpc and s390. > > It used to be so before offering archs to override the function for > tuning with commit ecd096506922 ("mm: make deferred init's max threads > arch-specific"). > > 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%) | > > Michael Ellerman on a powerpc machine (1TB, 40 cores, 4KB pages) reports > faster deferred_init_memmap from 210-240ms to 90-110ms between nodes. > > Signed-off-by: Eric Chanudet <echanude@redhat.com> > Tested-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Acked-by: Mike Rapoport (IBM) <rppt@kernel.org> > --- > - 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). > > - v2: https://lore.kernel.org/linux-arm-kernel/20240522203758.626932-4-echanude@redhat.com/ > - Changes since v2: > - deferred_page_init_max_threads returns unsigned and use max instead > of max_t. > - Make deferred_page_init_max_threads static since there are no more > override. > - Rephrase description. > - Add T-b and report from Michael Ellerman. > > arch/x86/mm/init_64.c | 12 ------------ > include/linux/memblock.h | 2 -- > mm/mm_init.c | 5 ++--- > 3 files changed, 2 insertions(+), 17 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/include/linux/memblock.h b/include/linux/memblock.h > index e2082240586d..40c62aca36ec 100644 > --- a/include/linux/memblock.h > +++ b/include/linux/memblock.h > @@ -335,8 +335,6 @@ void __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, > for (; i != U64_MAX; \ > __next_mem_pfn_range_in_zone(&i, zone, p_start, p_end)) > > -int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask); > - > #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ > > /** > diff --git a/mm/mm_init.c b/mm/mm_init.c > index f72b852bd5b8..acfeba508796 100644 > --- a/mm/mm_init.c > +++ b/mm/mm_init.c > @@ -2122,11 +2122,10 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, > } > } > > -/* An arch may override for more concurrency. */ > -__weak int __init > +static unsigned int __init > deferred_page_init_max_threads(const struct cpumask *node_cpumask) > { > - return 1; > + return max(cpumask_weight(node_cpumask), 1U); > } > > /* Initialise remaining memory on a node */ > -- > 2.44.0 >
On Tue, May 28, 2024 at 02:54:58PM -0400, Eric Chanudet wrote: > When DEFERRED_STRUCT_PAGE_INIT=y, use a node's cpu count as maximum > thread count for the deferred initialization of struct pages via padata. > This should result in shorter boot times for these configurations by > going through page_alloc_init_late() faster as systems tend not to be > under heavy load that early in the bootstrap. > > Only x86_64 does that now. Make it archs agnostic when > DEFERRED_STRUCT_PAGE_INIT is set. With the default defconfigs, that > includes powerpc and s390. > > It used to be so before offering archs to override the function for > tuning with commit ecd096506922 ("mm: make deferred init's max threads > arch-specific"). > > 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%) | > > Michael Ellerman on a powerpc machine (1TB, 40 cores, 4KB pages) reports > faster deferred_init_memmap from 210-240ms to 90-110ms between nodes. > > Signed-off-by: Eric Chanudet <echanude@redhat.com> > Tested-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) > > --- > - 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). > > - v2: https://lore.kernel.org/linux-arm-kernel/20240522203758.626932-4-echanude@redhat.com/ > - Changes since v2: > - deferred_page_init_max_threads returns unsigned and use max instead > of max_t. > - Make deferred_page_init_max_threads static since there are no more > override. > - Rephrase description. > - Add T-b and report from Michael Ellerman. > > arch/x86/mm/init_64.c | 12 ------------ > include/linux/memblock.h | 2 -- > mm/mm_init.c | 5 ++--- > 3 files changed, 2 insertions(+), 17 deletions(-) It does speed up. For s390: Acked-by: Alexander Gordeev <agordeev@linux.ibm.com>
On 05/28/24 at 02:54pm, Eric Chanudet wrote: > When DEFERRED_STRUCT_PAGE_INIT=y, use a node's cpu count as maximum > thread count for the deferred initialization of struct pages via padata. > This should result in shorter boot times for these configurations by > going through page_alloc_init_late() faster as systems tend not to be > under heavy load that early in the bootstrap. > > Only x86_64 does that now. Make it archs agnostic when > DEFERRED_STRUCT_PAGE_INIT is set. With the default defconfigs, that > includes powerpc and s390. > > It used to be so before offering archs to override the function for > tuning with commit ecd096506922 ("mm: make deferred init's max threads > arch-specific"). > > 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%) | > > Michael Ellerman on a powerpc machine (1TB, 40 cores, 4KB pages) reports > faster deferred_init_memmap from 210-240ms to 90-110ms between nodes. > > Signed-off-by: Eric Chanudet <echanude@redhat.com> > Tested-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) > > --- > - 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). > > - v2: https://lore.kernel.org/linux-arm-kernel/20240522203758.626932-4-echanude@redhat.com/ > - Changes since v2: > - deferred_page_init_max_threads returns unsigned and use max instead > of max_t. > - Make deferred_page_init_max_threads static since there are no more > override. > - Rephrase description. > - Add T-b and report from Michael Ellerman. > > arch/x86/mm/init_64.c | 12 ------------ > include/linux/memblock.h | 2 -- > mm/mm_init.c | 5 ++--- > 3 files changed, 2 insertions(+), 17 deletions(-) Reviewed-by: Baoquan He <bhe@redhat.com> > > 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/include/linux/memblock.h b/include/linux/memblock.h > index e2082240586d..40c62aca36ec 100644 > --- a/include/linux/memblock.h > +++ b/include/linux/memblock.h > @@ -335,8 +335,6 @@ void __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, > for (; i != U64_MAX; \ > __next_mem_pfn_range_in_zone(&i, zone, p_start, p_end)) > > -int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask); > - > #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ > > /** > diff --git a/mm/mm_init.c b/mm/mm_init.c > index f72b852bd5b8..acfeba508796 100644 > --- a/mm/mm_init.c > +++ b/mm/mm_init.c > @@ -2122,11 +2122,10 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, > } > } > > -/* An arch may override for more concurrency. */ > -__weak int __init > +static unsigned int __init > deferred_page_init_max_threads(const struct cpumask *node_cpumask) > { > - return 1; > + return max(cpumask_weight(node_cpumask), 1U); > } > > /* Initialise remaining memory on a node */ > -- > 2.44.0 >
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/include/linux/memblock.h b/include/linux/memblock.h index e2082240586d..40c62aca36ec 100644 --- a/include/linux/memblock.h +++ b/include/linux/memblock.h @@ -335,8 +335,6 @@ void __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone, for (; i != U64_MAX; \ __next_mem_pfn_range_in_zone(&i, zone, p_start, p_end)) -int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask); - #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */ /** diff --git a/mm/mm_init.c b/mm/mm_init.c index f72b852bd5b8..acfeba508796 100644 --- a/mm/mm_init.c +++ b/mm/mm_init.c @@ -2122,11 +2122,10 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn, } } -/* An arch may override for more concurrency. */ -__weak int __init +static unsigned int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask) { - return 1; + return max(cpumask_weight(node_cpumask), 1U); } /* Initialise remaining memory on a node */
When DEFERRED_STRUCT_PAGE_INIT=y, use a node's cpu count as maximum thread count for the deferred initialization of struct pages via padata. This should result in shorter boot times for these configurations by going through page_alloc_init_late() faster as systems tend not to be under heavy load that early in the bootstrap. Only x86_64 does that now. Make it archs agnostic when DEFERRED_STRUCT_PAGE_INIT is set. With the default defconfigs, that includes powerpc and s390. It used to be so before offering archs to override the function for tuning with commit ecd096506922 ("mm: make deferred init's max threads arch-specific"). 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%) | Michael Ellerman on a powerpc machine (1TB, 40 cores, 4KB pages) reports faster deferred_init_memmap from 210-240ms to 90-110ms between nodes. Signed-off-by: Eric Chanudet <echanude@redhat.com> Tested-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) --- - 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). - v2: https://lore.kernel.org/linux-arm-kernel/20240522203758.626932-4-echanude@redhat.com/ - Changes since v2: - deferred_page_init_max_threads returns unsigned and use max instead of max_t. - Make deferred_page_init_max_threads static since there are no more override. - Rephrase description. - Add T-b and report from Michael Ellerman. arch/x86/mm/init_64.c | 12 ------------ include/linux/memblock.h | 2 -- mm/mm_init.c | 5 ++--- 3 files changed, 2 insertions(+), 17 deletions(-)