| Message ID | 20230321170513.2401534-7-rppt@kernel.org (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm: move core MM initialization to mm/mm_init.c | expand |
On 3/21/23 18:05, Mike Rapoport wrote: > From: "Mike Rapoport (IBM)" <rppt@kernel.org> > > Both build_all_zonelists() and page_alloc_init_cpuhp() must be called > after SMP setup is complete but before the page allocator is set up. > > Still, they both are a part of memory management initialization, so move > them to mm_init(). Well, logic grouping is one thing, but not breaking a functional order is more important. So this moves both calls to happen later than theyw ere. I guess it could only matter for page_alloc_init_cpuhp() in case cpu hotplugs would be processed in some of the calls we "skipped" over by moving this later. And one of them is setup_arch()... so are we sure no arch does some cpu hotplug for non-boot cpus there? > Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org> > Acked-by: David Hildenbrand <david@redhat.com> > --- > init/main.c | 7 ++++--- > 1 file changed, 4 insertions(+), 3 deletions(-) > > diff --git a/init/main.c b/init/main.c > index b2499bee7a3c..4423906177c1 100644 > --- a/init/main.c > +++ b/init/main.c > @@ -833,6 +833,10 @@ static void __init report_meminit(void) > */ > static void __init mm_init(void) > { > + /* Initializations relying on SMP setup */ > + build_all_zonelists(NULL); > + page_alloc_init_cpuhp(); > + > /* > * page_ext requires contiguous pages, > * bigger than MAX_ORDER unless SPARSEMEM. > @@ -968,9 +972,6 @@ asmlinkage __visible void __init __no_sanitize_address start_kernel(void) > smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */ > boot_cpu_hotplug_init(); > > - build_all_zonelists(NULL); > - page_alloc_init_cpuhp(); > - > pr_notice("Kernel command line: %s\n", saved_command_line); > /* parameters may set static keys */ > jump_label_init();
On Wed, Mar 22, 2023 at 05:10:10PM +0100, Vlastimil Babka wrote: > On 3/21/23 18:05, Mike Rapoport wrote: > > From: "Mike Rapoport (IBM)" <rppt@kernel.org> > > > > Both build_all_zonelists() and page_alloc_init_cpuhp() must be called > > after SMP setup is complete but before the page allocator is set up. > > > > Still, they both are a part of memory management initialization, so move > > them to mm_init(). > > Well, logic grouping is one thing, but not breaking a functional order is > more important. So this moves both calls to happen later than theyw ere. I > guess it could only matter for page_alloc_init_cpuhp() in case cpu hotplugs > would be processed in some of the calls we "skipped" over by moving this > later. And one of them is setup_arch()... so are we sure no arch does some > cpu hotplug for non-boot cpus there? mm_init() happens after the point build_all_zonelists() and page_alloc_init_cpuhp() were originally, so they are essentially moved later in the init sequence and in either case called after setup_arch(). We skip the code below and it does not do neither cpu hotplug nor non-memblock allocations. jump_label_init(); parse_early_param(); after_dashes = parse_args("Booting kernel", static_command_line, __start___param, __stop___param - __start___param, -1, -1, NULL, &unknown_bootoption); print_unknown_bootoptions(); if (!IS_ERR_OR_NULL(after_dashes)) parse_args("Setting init args", after_dashes, NULL, 0, -1, -1, NULL, set_init_arg); if (extra_init_args) parse_args("Setting extra init args", extra_init_args, NULL, 0, -1, -1, NULL, set_init_arg); /* Architectural and non-timekeeping rng init, before allocator init */ random_init_early(command_line); /* * These use large bootmem allocations and must precede * kmem_cache_init() */ setup_log_buf(0); vfs_caches_init_early(); sort_main_extable(); trap_init(); > > Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org> > > Acked-by: David Hildenbrand <david@redhat.com> > > --- > > init/main.c | 7 ++++--- > > 1 file changed, 4 insertions(+), 3 deletions(-) > > > > diff --git a/init/main.c b/init/main.c > > index b2499bee7a3c..4423906177c1 100644 > > --- a/init/main.c > > +++ b/init/main.c > > @@ -833,6 +833,10 @@ static void __init report_meminit(void) > > */ > > static void __init mm_init(void) > > { > > + /* Initializations relying on SMP setup */ > > + build_all_zonelists(NULL); > > + page_alloc_init_cpuhp(); > > + > > /* > > * page_ext requires contiguous pages, > > * bigger than MAX_ORDER unless SPARSEMEM. > > @@ -968,9 +972,6 @@ asmlinkage __visible void __init __no_sanitize_address start_kernel(void) > > smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */ > > boot_cpu_hotplug_init(); > > > > - build_all_zonelists(NULL); > > - page_alloc_init_cpuhp(); > > - > > pr_notice("Kernel command line: %s\n", saved_command_line); > > /* parameters may set static keys */ > > jump_label_init(); >
On 3/22/23 21:26, Mike Rapoport wrote: > On Wed, Mar 22, 2023 at 05:10:10PM +0100, Vlastimil Babka wrote: >> On 3/21/23 18:05, Mike Rapoport wrote: >> > From: "Mike Rapoport (IBM)" <rppt@kernel.org> >> > >> > Both build_all_zonelists() and page_alloc_init_cpuhp() must be called >> > after SMP setup is complete but before the page allocator is set up. >> > >> > Still, they both are a part of memory management initialization, so move >> > them to mm_init(). >> >> Well, logic grouping is one thing, but not breaking a functional order is >> more important. So this moves both calls to happen later than theyw ere. I >> guess it could only matter for page_alloc_init_cpuhp() in case cpu hotplugs >> would be processed in some of the calls we "skipped" over by moving this >> later. And one of them is setup_arch()... so are we sure no arch does some >> cpu hotplug for non-boot cpus there? > > mm_init() happens after the point build_all_zonelists() and > page_alloc_init_cpuhp() were originally, so they are essentially moved > later in the init sequence and in either case called after setup_arch(). Right, I looked at a wrong place in start_kernel() for the original location of the calls, sorry for the noise. > We skip the code below and it does not do neither cpu hotplug nor > non-memblock allocations. > > jump_label_init(); > parse_early_param(); > after_dashes = parse_args("Booting kernel", > static_command_line, __start___param, > __stop___param - __start___param, > -1, -1, NULL, &unknown_bootoption); > print_unknown_bootoptions(); > if (!IS_ERR_OR_NULL(after_dashes)) > parse_args("Setting init args", after_dashes, NULL, 0, -1, -1, > NULL, set_init_arg); > if (extra_init_args) > parse_args("Setting extra init args", extra_init_args, > NULL, 0, -1, -1, NULL, set_init_arg); > > /* Architectural and non-timekeeping rng init, before allocator init */ > random_init_early(command_line); > > /* > * These use large bootmem allocations and must precede > * kmem_cache_init() > */ > setup_log_buf(0); > vfs_caches_init_early(); > sort_main_extable(); > trap_init(); > Yeah, that looks safe. >> > Signed-off-by: Mike Rapoport (IBM) <rppt@kernel.org> >> > Acked-by: David Hildenbrand <david@redhat.com> Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
diff --git a/init/main.c b/init/main.c index b2499bee7a3c..4423906177c1 100644 --- a/init/main.c +++ b/init/main.c @@ -833,6 +833,10 @@ static void __init report_meminit(void) */ static void __init mm_init(void) { + /* Initializations relying on SMP setup */ + build_all_zonelists(NULL); + page_alloc_init_cpuhp(); + /* * page_ext requires contiguous pages, * bigger than MAX_ORDER unless SPARSEMEM. @@ -968,9 +972,6 @@ asmlinkage __visible void __init __no_sanitize_address start_kernel(void) smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */ boot_cpu_hotplug_init(); - build_all_zonelists(NULL); - page_alloc_init_cpuhp(); - pr_notice("Kernel command line: %s\n", saved_command_line); /* parameters may set static keys */ jump_label_init();