.clang-format | 1 - include/linux/memblock.h | 22 ----- mm/memblock.c | 64 ------------- mm/mm_init.c | 195 +++++++++++++-------------------------- 4 files changed, 62 insertions(+), 220 deletions(-)
From: "Mike Rapoport (Microsoft)" <rppt@kernel.org> Hi, These patches simplify deferred initialization of the memory map. Beside nice negative diffstat I measured 3ms reduction in the initialization of deferred pages on single node system with 64GiB of RAM. I don't have access to large memory machines, so I'd really appreciate testing of these patches on them to make sure there's no regression there. The patches are also available at git: https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git/log/?h=deferred-memmap-init/v1 Mike Rapoport (Microsoft) (4): mm/mm_init: use deferred_init_memmap_chunk() in deferred_grow_zone() mm/mm_init: deferred_init_memmap: use a job per zone mm/mm_init: drop deferred_init_maxorder() memblock: drop for_each_free_mem_pfn_range_in_zone_from() .clang-format | 1 - include/linux/memblock.h | 22 ----- mm/memblock.c | 64 ------------- mm/mm_init.c | 195 +++++++++++++-------------------------- 4 files changed, 62 insertions(+), 220 deletions(-) base-commit: 8f5ae30d69d7543eee0d70083daf4de8fe15d585 -- 2.50.1
On Mon, Aug 18, 2025 at 09:46:11AM +0300, Mike Rapoport wrote: > From: "Mike Rapoport (Microsoft)" <rppt@kernel.org> > > Hi, > > These patches simplify deferred initialization of the memory map. > > Beside nice negative diffstat I measured 3ms reduction in the > initialization of deferred pages on single node system with 64GiB of RAM. > > I don't have access to large memory machines, so I'd really appreciate > testing of these patches on them to make sure there's no regression there. > > The patches are also available at git: > https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git/log/?h=deferred-memmap-init/v1 > > Mike Rapoport (Microsoft) (4): > mm/mm_init: use deferred_init_memmap_chunk() in deferred_grow_zone() > mm/mm_init: deferred_init_memmap: use a job per zone > mm/mm_init: drop deferred_init_maxorder() > memblock: drop for_each_free_mem_pfn_range_in_zone_from() I've added those to memblock tree > .clang-format | 1 - > include/linux/memblock.h | 22 ----- > mm/memblock.c | 64 ------------- > mm/mm_init.c | 195 +++++++++++++-------------------------- > 4 files changed, 62 insertions(+), 220 deletions(-) > > > base-commit: 8f5ae30d69d7543eee0d70083daf4de8fe15d585 > -- > 2.50.1 > -- Sincerely yours, Mike.
On Mon, Aug 18, 2025 at 09:46:11AM +0300, Mike Rapoport wrote: >From: "Mike Rapoport (Microsoft)" <rppt@kernel.org> > >Hi, > >These patches simplify deferred initialization of the memory map. > >Beside nice negative diffstat I measured 3ms reduction in the >initialization of deferred pages on single node system with 64GiB of RAM. Nice cleanup. For this series: Reviewed-by: Wei Yang <richard.weiyang@gmail.com> I guess the speed up is from "use a job per zone". So we do initialization per zone instead of per memblock range in the zone, right? > >I don't have access to large memory machines, so I'd really appreciate >testing of these patches on them to make sure there's no regression there. > >The patches are also available at git: >https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git/log/?h=deferred-memmap-init/v1 > >Mike Rapoport (Microsoft) (4): > mm/mm_init: use deferred_init_memmap_chunk() in deferred_grow_zone() > mm/mm_init: deferred_init_memmap: use a job per zone > mm/mm_init: drop deferred_init_maxorder() > memblock: drop for_each_free_mem_pfn_range_in_zone_from() > > .clang-format | 1 - > include/linux/memblock.h | 22 ----- > mm/memblock.c | 64 ------------- > mm/mm_init.c | 195 +++++++++++++-------------------------- > 4 files changed, 62 insertions(+), 220 deletions(-) > > >base-commit: 8f5ae30d69d7543eee0d70083daf4de8fe15d585 >-- >2.50.1 > -- Wei Yang Help you, Help me
On Tue, Aug 19, 2025 at 07:39:41AM +0000, Wei Yang wrote: > On Mon, Aug 18, 2025 at 09:46:11AM +0300, Mike Rapoport wrote: > >From: "Mike Rapoport (Microsoft)" <rppt@kernel.org> > > > >Hi, > > > >These patches simplify deferred initialization of the memory map. > > > >Beside nice negative diffstat I measured 3ms reduction in the > >initialization of deferred pages on single node system with 64GiB of RAM. > > Nice cleanup. > > For this series: > > Reviewed-by: Wei Yang <richard.weiyang@gmail.com> Thanks! > I guess the speed up is from "use a job per zone". So we do initialization per > zone instead of per memblock range in the zone, right? Yes, we run a job per zone instead of a job per memblock range in the zone. > >I don't have access to large memory machines, so I'd really appreciate > >testing of these patches on them to make sure there's no regression there. > > > >The patches are also available at git: > >https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git/log/?h=deferred-memmap-init/v1 > > > >Mike Rapoport (Microsoft) (4): > > mm/mm_init: use deferred_init_memmap_chunk() in deferred_grow_zone() > > mm/mm_init: deferred_init_memmap: use a job per zone > > mm/mm_init: drop deferred_init_maxorder() > > memblock: drop for_each_free_mem_pfn_range_in_zone_from() > > > > .clang-format | 1 - > > include/linux/memblock.h | 22 ----- > > mm/memblock.c | 64 ------------- > > mm/mm_init.c | 195 +++++++++++++-------------------------- > > 4 files changed, 62 insertions(+), 220 deletions(-) > > > > > >base-commit: 8f5ae30d69d7543eee0d70083daf4de8fe15d585 > >-- > >2.50.1 > > > > -- > Wei Yang > Help you, Help me -- Sincerely yours, Mike.
© 2016 - 2025 Red Hat, Inc.