Pmem September 2022

pmem@lists.openanolis.cn

2 参与者
3 讨论

Re: [PATCH v2] mm: make cpu-less memory node to movable node only

by Baolin Wang

On 9/1/2022 3:46 PM, Feng Tang wrote: > From 1784d7c7a6911a3d6a93d1847d84e9297c8e4f4c Mon Sep 17 00:00:00 2001 > From: Feng Tang <feng.tang(a)intel.com> > Date: Wed, 31 Aug 2022 15:31:04 +0800 > Subject: [PATCH v2] mm: make cpu-less memory node to movable node only > > When platform has PMEM node, those node are mostly a pure memory node > without any CPU attached. To avoid performance/latency sensitive > kernel allocation on these nodes, make all cpu-less memory nodes > be enabled as node with movable zone only. > > And when users want it to be normal node, appending "cpuless_node_normal" > to kernel cmdline parameters can achive that. > > Signed-off-by: Feng Tang <feng.tang(a)intel.com> > Signed-off-by: Huaisheng Ye <huaisheng.ye(a)intel.com> LGTM. Thanks. Reviewed-by: Baolin Wang <baolin.wang(a)linux.alibaba.com> > --- > Changelog: > > Since v1 > * directly jump to out2 after movable handling (Baolin Wang) > > arch/x86/mm/srat.c | 5 +++++ > drivers/base/node.c | 2 ++ > include/linux/nodemask.h | 1 + > mm/page_alloc.c | 32 ++++++++++++++++++++++++++++++++ > 4 files changed, 40 insertions(+) > > diff --git a/arch/x86/mm/srat.c b/arch/x86/mm/srat.c > index dac07e4f5834..dcb568067d79 100644 > --- a/arch/x86/mm/srat.c > +++ b/arch/x86/mm/srat.c > @@ -60,6 +60,9 @@ acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa) > node_set(node, numa_nodes_parsed); > printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%04x -> Node %u\n", > pxm, apic_id, node); > + > + if (!node_state(node, N_POSSIBLE_CPU)) > + node_set_state(node, N_POSSIBLE_CPU); > } > > /* Callback for Proximity Domain -> LAPIC mapping */ > @@ -101,6 +104,8 @@ acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa) > node_set(node, numa_nodes_parsed); > printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%02x -> Node %u\n", > pxm, apic_id, node); > + if (!node_state(node, N_POSSIBLE_CPU)) > + node_set_state(node, N_POSSIBLE_CPU); > } > > int __init x86_acpi_numa_init(void) > diff --git a/drivers/base/node.c b/drivers/base/node.c > index 619fb1e38dd2..39ee52c69609 100644 > --- a/drivers/base/node.c > +++ b/drivers/base/node.c > @@ -990,6 +990,7 @@ static struct node_attr node_state_attr[] = { > #endif > [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), > [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), > + [N_POSSIBLE_CPU] = _NODE_ATTR(has_possible_cpu, N_POSSIBLE_CPU), > }; > > static struct attribute *node_state_attrs[] = { > @@ -1001,6 +1002,7 @@ static struct attribute *node_state_attrs[] = { > #endif > &node_state_attr[N_MEMORY].attr.attr, > &node_state_attr[N_CPU].attr.attr, > + &node_state_attr[N_POSSIBLE_CPU].attr.attr, > NULL > }; > > diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h > index 5a30ad594ccc..ad3100b06938 100644 > --- a/include/linux/nodemask.h > +++ b/include/linux/nodemask.h > @@ -399,6 +399,7 @@ enum node_states { > #endif > N_MEMORY, /* The node has memory(regular, high, movable) */ > N_CPU, /* The node has one or more cpus */ > + N_POSSIBLE_CPU, /* possible nodes that have cpu attached, only used in early boot phase */ > NR_NODE_STATES > }; > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 8cfce927fb9b..f20976536cfc 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -294,6 +294,7 @@ static unsigned long required_kernelcore_percent __initdata; > static unsigned long required_movablecore __initdata; > static unsigned long required_movablecore_percent __initdata; > static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata; > +static bool enable_cpuless_memnode_normal_node __initdata; > static bool mirrored_kernelcore __meminitdata; > > /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ > @@ -7163,6 +7164,26 @@ static void __init find_zone_movable_pfns_for_nodes(void) > required_movablecore = (totalpages * 100 * required_movablecore_percent) / > 10000UL; > > + if (!required_kernelcore && !required_movablecore && > + !nodes_empty(node_states[N_POSSIBLE_CPU]) && > + !enable_cpuless_memnode_normal_node) { > + > + /* Put meory from cpu-less nodes into movable zones */ > + for_each_memblock(memory, r) { > + nid = memblock_get_region_node(r); > + > + if (node_isset(nid, node_states[N_POSSIBLE_CPU])) > + continue; > + > + usable_startpfn = PFN_DOWN(r->base); > + zone_movable_pfn[nid] = zone_movable_pfn[nid] ? > + min(usable_startpfn, zone_movable_pfn[nid]) : > + usable_startpfn; > + } > + > + goto out2; > + } > + > /* > * If movablecore= was specified, calculate what size of > * kernelcore that corresponds so that memory usable for > @@ -7457,8 +7478,19 @@ static int __init cmdline_parse_movablecore(char *p) > &required_movablecore_percent); > } > > +/* > + * cpuless memory nodes will be enabled to movable node by default, > + * add this cmdline to make it be enabled as a normal node > + */ > +static int __init cmdline_parse_cpuless_memnode(char *str) > +{ > + enable_cpuless_memnode_normal_node = true; > + return 0; > +} > + > early_param("kernelcore", cmdline_parse_kernelcore); > early_param("movablecore", cmdline_parse_movablecore); > +early_param("cpuless_node_normal", cmdline_parse_cpuless_memnode); > > #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ >

2 years, 3 months

Re: [PATCH] mm: make cpu-less memory node to movable node only

by zhong jiang

On 2022/9/1 9:46 上午, Feng Tang wrote: > Hi Zhong Jiang, > > Thanks for the review! > > On Wed, Aug 31, 2022 at 07:20:21PM +0800, zhong jiang wrote: >> On 2022/8/31 3:31 下午, Huaisheng Ye wrote: >>> From: Feng Tang <feng.tang(a)intel.com> >>> >>> When platform has PMEM node, those node are mostly a pure memory node >>> without any CPU attached. To avoid performance/latency sensitive >>> kernel allocation on these nodes, make all cpu-less memory nodes >>> be enabled as node with movable zone only. >>> >>> And when users want it to be normal node, appending "cpuless_node_normal" >>> to kernel cmdline parameters can achive that. >>> >>> Signed-off-by: Feng Tang <feng.tang(a)intel.com> >>> Signed-off-by: Huaisheng Ye <huaisheng.ye(a)intel.com> >>> --- >>> arch/x86/mm/srat.c | 5 +++++ >>> drivers/base/node.c | 2 ++ >>> include/linux/nodemask.h | 1 + >>> mm/page_alloc.c | 30 ++++++++++++++++++++++++++++++ >>> 4 files changed, 38 insertions(+) >>> >>> diff --git a/arch/x86/mm/srat.c b/arch/x86/mm/srat.c >>> index dac07e4f5834..dcb568067d79 100644 >>> --- a/arch/x86/mm/srat.c >>> +++ b/arch/x86/mm/srat.c >>> @@ -60,6 +60,9 @@ acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa) >>> node_set(node, numa_nodes_parsed); >>> printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%04x -> Node %u\n", >>> pxm, apic_id, node); >>> + >>> + if (!node_state(node, N_POSSIBLE_CPU)) >>> + node_set_state(node, N_POSSIBLE_CPU); >>> } >>> /* Callback for Proximity Domain -> LAPIC mapping */ >>> @@ -101,6 +104,8 @@ acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa) >>> node_set(node, numa_nodes_parsed); >>> printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%02x -> Node %u\n", >>> pxm, apic_id, node); >>> + if (!node_state(node, N_POSSIBLE_CPU)) >>> + node_set_state(node, N_POSSIBLE_CPU); >>> } >> IIUC, pmem memory also store in the srat table, the above change seems to >> fail to work because pmem memory has marked as the N_POSSIBLE_CPU. > PMEM node will not be set for N_POSSIBLE_CPU, the upper 2 functions > are only for parsing CPU entry with interrupt controllers. > > IIUC, you are refering the NUMA node info maintained by architectural > code (say arch/x86): node_possible_map / numa_nodes_parsed, where PMEM > node will be set. > > And the node map here N_POSSIBLE_CPU is from another set mainly for > mm/ code, and used later in memory init part. > > This patch has been tested on QEMU and RE7P machines. Thanks for your clarification. > Thanks, > Feng > > >> Am I missing? >> >> Thanks, >>> int __init x86_acpi_numa_init(void) >>> diff --git a/drivers/base/node.c b/drivers/base/node.c >>> index 619fb1e38dd2..39ee52c69609 100644 >>> --- a/drivers/base/node.c >>> +++ b/drivers/base/node.c >>> @@ -990,6 +990,7 @@ static struct node_attr node_state_attr[] = { >>> #endif >>> [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), >>> [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), >>> + [N_POSSIBLE_CPU] = _NODE_ATTR(has_possible_cpu, N_POSSIBLE_CPU), >>> }; >>> static struct attribute *node_state_attrs[] = { >>> @@ -1001,6 +1002,7 @@ static struct attribute *node_state_attrs[] = { >>> #endif >>> &node_state_attr[N_MEMORY].attr.attr, >>> &node_state_attr[N_CPU].attr.attr, >>> + &node_state_attr[N_POSSIBLE_CPU].attr.attr, >>> NULL >>> }; >>> diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h >>> index 5a30ad594ccc..ad3100b06938 100644 >>> --- a/include/linux/nodemask.h >>> +++ b/include/linux/nodemask.h >>> @@ -399,6 +399,7 @@ enum node_states { >>> #endif >>> N_MEMORY, /* The node has memory(regular, high, movable) */ >>> N_CPU, /* The node has one or more cpus */ >>> + N_POSSIBLE_CPU, /* possible nodes that have cpu attached, only used in early boot phase */ >>> NR_NODE_STATES >>> }; >>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c >>> index 8cfce927fb9b..9570ee499c12 100644 >>> --- a/mm/page_alloc.c >>> +++ b/mm/page_alloc.c >>> @@ -294,6 +294,7 @@ static unsigned long required_kernelcore_percent __initdata; >>> static unsigned long required_movablecore __initdata; >>> static unsigned long required_movablecore_percent __initdata; >>> static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata; >>> +static bool enable_cpuless_memnode_normal_node __initdata; >>> static bool mirrored_kernelcore __meminitdata; >>> /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ >>> @@ -7163,6 +7164,24 @@ static void __init find_zone_movable_pfns_for_nodes(void) >>> required_movablecore = (totalpages * 100 * required_movablecore_percent) / >>> 10000UL; >>> + if (!required_kernelcore && !required_movablecore && >>> + !nodes_empty(node_states[N_POSSIBLE_CPU]) && >>> + !enable_cpuless_memnode_normal_node) { >>> + >>> + /* Put meory from cpu-less nodes into movable zones */ >>> + for_each_memblock(memory, r) { >>> + nid = memblock_get_region_node(r); >>> + >>> + if (node_isset(nid, node_states[N_POSSIBLE_CPU])) >>> + continue; >>> + >>> + usable_startpfn = PFN_DOWN(r->base); >>> + zone_movable_pfn[nid] = zone_movable_pfn[nid] ? >>> + min(usable_startpfn, zone_movable_pfn[nid]) : >>> + usable_startpfn; >>> + } >>> + } >>> + >>> /* >>> * If movablecore= was specified, calculate what size of >>> * kernelcore that corresponds so that memory usable for >>> @@ -7457,8 +7476,19 @@ static int __init cmdline_parse_movablecore(char *p) >>> &required_movablecore_percent); >>> } >>> +/* >>> + * cpuless memory nodes will be enabled to movable node by default, >>> + * add this cmdline to make it be enabled as a normal node >>> + */ >>> +static int __init cmdline_parse_cpuless_memnode(char *str) >>> +{ >>> + enable_cpuless_memnode_normal_node = true; >>> + return 0; >>> +} >>> + >>> early_param("kernelcore", cmdline_parse_kernelcore); >>> early_param("movablecore", cmdline_parse_movablecore); >>> +early_param("cpuless_node_normal", cmdline_parse_cpuless_memnode); >>> #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */

2 years, 3 months

[PATCH] mm: make cpu-less memory node to movable node only

by Huaisheng Ye

From: Feng Tang <feng.tang(a)intel.com> When platform has PMEM node, those node are mostly a pure memory node without any CPU attached. To avoid performance/latency sensitive kernel allocation on these nodes, make all cpu-less memory nodes be enabled as node with movable zone only. And when users want it to be normal node, appending "cpuless_node_normal" to kernel cmdline parameters can achive that. Signed-off-by: Feng Tang <feng.tang(a)intel.com> Signed-off-by: Huaisheng Ye <huaisheng.ye(a)intel.com> --- arch/x86/mm/srat.c | 5 +++++ drivers/base/node.c | 2 ++ include/linux/nodemask.h | 1 + mm/page_alloc.c | 30 ++++++++++++++++++++++++++++++ 4 files changed, 38 insertions(+) diff --git a/arch/x86/mm/srat.c b/arch/x86/mm/srat.c index dac07e4f5834..dcb568067d79 100644 --- a/arch/x86/mm/srat.c +++ b/arch/x86/mm/srat.c @@ -60,6 +60,9 @@ acpi_numa_x2apic_affinity_init(struct acpi_srat_x2apic_cpu_affinity *pa) node_set(node, numa_nodes_parsed); printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%04x -> Node %u\n", pxm, apic_id, node); + + if (!node_state(node, N_POSSIBLE_CPU)) + node_set_state(node, N_POSSIBLE_CPU); } /* Callback for Proximity Domain -> LAPIC mapping */ @@ -101,6 +104,8 @@ acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa) node_set(node, numa_nodes_parsed); printk(KERN_INFO "SRAT: PXM %u -> APIC 0x%02x -> Node %u\n", pxm, apic_id, node); + if (!node_state(node, N_POSSIBLE_CPU)) + node_set_state(node, N_POSSIBLE_CPU); } int __init x86_acpi_numa_init(void) diff --git a/drivers/base/node.c b/drivers/base/node.c index 619fb1e38dd2..39ee52c69609 100644 --- a/drivers/base/node.c +++ b/drivers/base/node.c @@ -990,6 +990,7 @@ static struct node_attr node_state_attr[] = { #endif [N_MEMORY] = _NODE_ATTR(has_memory, N_MEMORY), [N_CPU] = _NODE_ATTR(has_cpu, N_CPU), + [N_POSSIBLE_CPU] = _NODE_ATTR(has_possible_cpu, N_POSSIBLE_CPU), }; static struct attribute *node_state_attrs[] = { @@ -1001,6 +1002,7 @@ static struct attribute *node_state_attrs[] = { #endif &node_state_attr[N_MEMORY].attr.attr, &node_state_attr[N_CPU].attr.attr, + &node_state_attr[N_POSSIBLE_CPU].attr.attr, NULL }; diff --git a/include/linux/nodemask.h b/include/linux/nodemask.h index 5a30ad594ccc..ad3100b06938 100644 --- a/include/linux/nodemask.h +++ b/include/linux/nodemask.h @@ -399,6 +399,7 @@ enum node_states { #endif N_MEMORY, /* The node has memory(regular, high, movable) */ N_CPU, /* The node has one or more cpus */ + N_POSSIBLE_CPU, /* possible nodes that have cpu attached, only used in early boot phase */ NR_NODE_STATES }; diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 8cfce927fb9b..9570ee499c12 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -294,6 +294,7 @@ static unsigned long required_kernelcore_percent __initdata; static unsigned long required_movablecore __initdata; static unsigned long required_movablecore_percent __initdata; static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata; +static bool enable_cpuless_memnode_normal_node __initdata; static bool mirrored_kernelcore __meminitdata; /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ @@ -7163,6 +7164,24 @@ static void __init find_zone_movable_pfns_for_nodes(void) required_movablecore = (totalpages * 100 * required_movablecore_percent) / 10000UL; + if (!required_kernelcore && !required_movablecore && + !nodes_empty(node_states[N_POSSIBLE_CPU]) && + !enable_cpuless_memnode_normal_node) { + + /* Put meory from cpu-less nodes into movable zones */ + for_each_memblock(memory, r) { + nid = memblock_get_region_node(r); + + if (node_isset(nid, node_states[N_POSSIBLE_CPU])) + continue; + + usable_startpfn = PFN_DOWN(r->base); + zone_movable_pfn[nid] = zone_movable_pfn[nid] ? + min(usable_startpfn, zone_movable_pfn[nid]) : + usable_startpfn; + } + } + /* * If movablecore= was specified, calculate what size of * kernelcore that corresponds so that memory usable for @@ -7457,8 +7476,19 @@ static int __init cmdline_parse_movablecore(char *p) &required_movablecore_percent); } +/* + * cpuless memory nodes will be enabled to movable node by default, + * add this cmdline to make it be enabled as a normal node + */ +static int __init cmdline_parse_cpuless_memnode(char *str) +{ + enable_cpuless_memnode_normal_node = true; + return 0; +} + early_param("kernelcore", cmdline_parse_kernelcore); early_param("movablecore", cmdline_parse_movablecore); +early_param("cpuless_node_normal", cmdline_parse_cpuless_memnode); #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ -- 2.27.0

2 years, 3 months

2024

2023

2022

2021

Pmem September 2022