| Message ID | 157966229575.2508551.1892426244277171485.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Memory Hierarchy: Enable target node lookups for reserved memory | expand |
* Dan Williams <dan.j.williams@intel.com> wrote: > Currently x86 numa_meminfo is marked __initdata in the > CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow > drivers to map reserved memory to a 'target_node' > (phys_to_target_node()), add support for removing the __initdata > designation for those users. Both memory hotplug and > phys_to_target_node() users select CONFIG_KEEP_NUMA to tell the arch to > maintain its physical address to numa mapping infrastructure post init. > > Cc: Dave Hansen <dave.hansen@linux.intel.com> > Cc: Andy Lutomirski <luto@kernel.org> > Cc: Peter Zijlstra <peterz@infradead.org> > Cc: Thomas Gleixner <tglx@linutronix.de> > Cc: Ingo Molnar <mingo@redhat.com> > Cc: Borislav Petkov <bp@alien8.de> > Cc: "H. Peter Anvin" <hpa@zytor.com> > Cc: <x86@kernel.org> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: David Hildenbrand <david@redhat.com> > Cc: Michal Hocko <mhocko@suse.com> > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > --- > arch/x86/mm/numa.c | 6 +----- > include/linux/numa.h | 6 ++++++ > mm/Kconfig | 5 +++++ > 3 files changed, 12 insertions(+), 5 deletions(-) The concept and the x86 portions look sane, just a few minor nits: > > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c > index 99f7a68738f0..5289d9d6799a 100644 > --- a/arch/x86/mm/numa.c > +++ b/arch/x86/mm/numa.c > @@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata; > struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; > EXPORT_SYMBOL(node_data); > > -static struct numa_meminfo numa_meminfo > -#ifndef CONFIG_MEMORY_HOTPLUG > -__initdata > -#endif > -; > +static struct numa_meminfo numa_meminfo __initdata_numa; > > static int numa_distance_cnt; > static u8 *numa_distance; > diff --git a/include/linux/numa.h b/include/linux/numa.h > index 20f4e44b186c..c005ed6b807b 100644 > --- a/include/linux/numa.h > +++ b/include/linux/numa.h > @@ -13,6 +13,12 @@ > > #define NUMA_NO_NODE (-1) > > +#ifdef CONFIG_KEEP_NUMA > +#define __initdata_numa > +#else > +#define __initdata_numa __initdata > +#endif > + > #ifdef CONFIG_NUMA > int numa_map_to_online_node(int node); > #else > diff --git a/mm/Kconfig b/mm/Kconfig > index ab80933be65f..001f1185eadf 100644 > --- a/mm/Kconfig > +++ b/mm/Kconfig > @@ -139,6 +139,10 @@ config HAVE_FAST_GUP > config ARCH_KEEP_MEMBLOCK > bool > > +# Keep arch numa mapping infrastructure post-init. s/numa/NUMA Please also capitalize consistently in the rest of the series. > +config KEEP_NUMA > + bool So most of our recent new NUMA options followed the naming pattern of: CONFIG_NUMA_* Such as CONFIG_NUMA_BALANCING or CONFIG_NUMA_EMU. So I'd suggesting naming it to CONFIG_NUMA_KEEP, or, a bit more descriptively, such as CONFIG_NUMA_KEEP_MAPPING or such? 'Keeping NUMA' is kind of lame - of course we keep NUMA. ;-) Thanks, Ingo
Dan Williams <dan.j.williams@intel.com> writes: > +#ifdef CONFIG_KEEP_NUMA > +#define __initdata_numa > +#else > +#define __initdata_numa __initdata > +#endif TBH, I find this conditional annotation mightingly confusing. __initdata_numa still suggest that this is __initdata, just a different section and some extra rules or whatever. Something like __initdata_or_keepnuma (sorry I could not come up with something prettier, but you get the idea. Thanks, tglx
On Thu, Feb 13, 2020 at 1:32 AM Ingo Molnar <mingo@kernel.org> wrote: > > > * Dan Williams <dan.j.williams@intel.com> wrote: > > > Currently x86 numa_meminfo is marked __initdata in the > > CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow > > drivers to map reserved memory to a 'target_node' > > (phys_to_target_node()), add support for removing the __initdata > > designation for those users. Both memory hotplug and > > phys_to_target_node() users select CONFIG_KEEP_NUMA to tell the arch to > > maintain its physical address to numa mapping infrastructure post init. > > > > Cc: Dave Hansen <dave.hansen@linux.intel.com> > > Cc: Andy Lutomirski <luto@kernel.org> > > Cc: Peter Zijlstra <peterz@infradead.org> > > Cc: Thomas Gleixner <tglx@linutronix.de> > > Cc: Ingo Molnar <mingo@redhat.com> > > Cc: Borislav Petkov <bp@alien8.de> > > Cc: "H. Peter Anvin" <hpa@zytor.com> > > Cc: <x86@kernel.org> > > Cc: Andrew Morton <akpm@linux-foundation.org> > > Cc: David Hildenbrand <david@redhat.com> > > Cc: Michal Hocko <mhocko@suse.com> > > Signed-off-by: Dan Williams <dan.j.williams@intel.com> > > --- > > arch/x86/mm/numa.c | 6 +----- > > include/linux/numa.h | 6 ++++++ > > mm/Kconfig | 5 +++++ > > 3 files changed, 12 insertions(+), 5 deletions(-) > > The concept and the x86 portions look sane, just a few minor nits: > > > > > diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c > > index 99f7a68738f0..5289d9d6799a 100644 > > --- a/arch/x86/mm/numa.c > > +++ b/arch/x86/mm/numa.c > > @@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata; > > struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; > > EXPORT_SYMBOL(node_data); > > > > -static struct numa_meminfo numa_meminfo > > -#ifndef CONFIG_MEMORY_HOTPLUG > > -__initdata > > -#endif > > -; > > +static struct numa_meminfo numa_meminfo __initdata_numa; > > > > static int numa_distance_cnt; > > static u8 *numa_distance; > > diff --git a/include/linux/numa.h b/include/linux/numa.h > > index 20f4e44b186c..c005ed6b807b 100644 > > --- a/include/linux/numa.h > > +++ b/include/linux/numa.h > > @@ -13,6 +13,12 @@ > > > > #define NUMA_NO_NODE (-1) > > > > +#ifdef CONFIG_KEEP_NUMA > > +#define __initdata_numa > > +#else > > +#define __initdata_numa __initdata > > +#endif > > + > > #ifdef CONFIG_NUMA > > int numa_map_to_online_node(int node); > > #else > > diff --git a/mm/Kconfig b/mm/Kconfig > > index ab80933be65f..001f1185eadf 100644 > > --- a/mm/Kconfig > > +++ b/mm/Kconfig > > @@ -139,6 +139,10 @@ config HAVE_FAST_GUP > > config ARCH_KEEP_MEMBLOCK > > bool > > > > +# Keep arch numa mapping infrastructure post-init. > > s/numa/NUMA > > Please also capitalize consistently in the rest of the series. > > > +config KEEP_NUMA > > + bool > > > So most of our recent new NUMA options followed the naming pattern of: > > CONFIG_NUMA_* > > Such as CONFIG_NUMA_BALANCING or CONFIG_NUMA_EMU. > > So I'd suggesting naming it to CONFIG_NUMA_KEEP, or, a bit more > descriptively, such as CONFIG_NUMA_KEEP_MAPPING or such? > > 'Keeping NUMA' is kind of lame - of course we keep NUMA. ;-) Ok, I settled on CONFIG_NUMA_KEEP_MEMINFO, and will fix up the lowercase "numa" instances in the set.
On Thu, Feb 13, 2020 at 3:22 AM Thomas Gleixner <tglx@linutronix.de> wrote: > > Dan Williams <dan.j.williams@intel.com> writes: > > +#ifdef CONFIG_KEEP_NUMA > > +#define __initdata_numa > > +#else > > +#define __initdata_numa __initdata > > +#endif > > TBH, I find this conditional annotation mightingly confusing. > > __initdata_numa still suggest that this is __initdata, just a different > section and some extra rules or whatever. > > Something like __initdata_or_keepnuma (sorry I could not come up with > something prettier, but you get the idea. Yes, and to dovetail with Ingo's feedback I think __initdata_or_meminfo conveys it's optionally init vs runtime data.
diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c index 99f7a68738f0..5289d9d6799a 100644 --- a/arch/x86/mm/numa.c +++ b/arch/x86/mm/numa.c @@ -25,11 +25,7 @@ nodemask_t numa_nodes_parsed __initdata; struct pglist_data *node_data[MAX_NUMNODES] __read_mostly; EXPORT_SYMBOL(node_data); -static struct numa_meminfo numa_meminfo -#ifndef CONFIG_MEMORY_HOTPLUG -__initdata -#endif -; +static struct numa_meminfo numa_meminfo __initdata_numa; static int numa_distance_cnt; static u8 *numa_distance; diff --git a/include/linux/numa.h b/include/linux/numa.h index 20f4e44b186c..c005ed6b807b 100644 --- a/include/linux/numa.h +++ b/include/linux/numa.h @@ -13,6 +13,12 @@ #define NUMA_NO_NODE (-1) +#ifdef CONFIG_KEEP_NUMA +#define __initdata_numa +#else +#define __initdata_numa __initdata +#endif + #ifdef CONFIG_NUMA int numa_map_to_online_node(int node); #else diff --git a/mm/Kconfig b/mm/Kconfig index ab80933be65f..001f1185eadf 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -139,6 +139,10 @@ config HAVE_FAST_GUP config ARCH_KEEP_MEMBLOCK bool +# Keep arch numa mapping infrastructure post-init. +config KEEP_NUMA + bool + config MEMORY_ISOLATION bool @@ -154,6 +158,7 @@ config MEMORY_HOTPLUG bool "Allow for memory hot-add" depends on SPARSEMEM || X86_64_ACPI_NUMA depends on ARCH_ENABLE_MEMORY_HOTPLUG + select KEEP_NUMA if NUMA config MEMORY_HOTPLUG_SPARSE def_bool y
Currently x86 numa_meminfo is marked __initdata in the CONFIG_MEMORY_HOTPLUG=n case. In support of a new facility to allow drivers to map reserved memory to a 'target_node' (phys_to_target_node()), add support for removing the __initdata designation for those users. Both memory hotplug and phys_to_target_node() users select CONFIG_KEEP_NUMA to tell the arch to maintain its physical address to numa mapping infrastructure post init. Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: <x86@kernel.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Hildenbrand <david@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- arch/x86/mm/numa.c | 6 +----- include/linux/numa.h | 6 ++++++ mm/Kconfig | 5 +++++ 3 files changed, 12 insertions(+), 5 deletions(-)