From eabd9db64ea8ba64d2a0b1d70da38e1a95dcd08b Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Thu, 13 Jun 2024 16:54:33 +0800 Subject: ACPI: RISCV: Add NUMA support based on SRAT and SLIT Add acpi_numa.c file to enable parse NUMA information from ACPI SRAT and SLIT tables. SRAT table provide CPUs(Hart) and memory nodes to proximity domain mapping, while SLIT table provide the distance metrics between proximity domains. Signed-off-by: Haibo Xu Reviewed-by: Sunil V L Reviewed-by: Hanjun Guo Link: https://lore.kernel.org/r/65dbad1fda08a32922c44886e4581e49b4a2fecc.1718268003.git.haibo1.xu@intel.com Signed-off-by: Palmer Dabbelt --- arch/riscv/include/asm/acpi.h | 15 ++++- arch/riscv/kernel/Makefile | 1 + arch/riscv/kernel/acpi.c | 5 -- arch/riscv/kernel/acpi_numa.c | 131 ++++++++++++++++++++++++++++++++++++++++++ arch/riscv/kernel/setup.c | 4 +- arch/riscv/kernel/smpboot.c | 2 - include/linux/acpi.h | 6 ++ 7 files changed, 154 insertions(+), 10 deletions(-) create mode 100644 arch/riscv/kernel/acpi_numa.c diff --git a/arch/riscv/include/asm/acpi.h b/arch/riscv/include/asm/acpi.h index 7dad0cf9d701..e0a1f84404f3 100644 --- a/arch/riscv/include/asm/acpi.h +++ b/arch/riscv/include/asm/acpi.h @@ -61,11 +61,14 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { } void acpi_init_rintc_map(void); struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu); -u32 get_acpi_id_for_cpu(int cpu); +static inline u32 get_acpi_id_for_cpu(int cpu) +{ + return acpi_cpu_get_madt_rintc(cpu)->uid; +} + int acpi_get_riscv_isa(struct acpi_table_header *table, unsigned int cpu, const char **isa); -static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; } void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size, u32 *cboz_size, u32 *cbop_size); #else @@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct acpi_table_header *table, #endif /* CONFIG_ACPI */ +#ifdef CONFIG_ACPI_NUMA +int acpi_numa_get_nid(unsigned int cpu); +void acpi_map_cpus_to_nodes(void); +#else +static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; } +static inline void acpi_map_cpus_to_nodes(void) { } +#endif /* CONFIG_ACPI_NUMA */ + #endif /*_ASM_ACPI_H*/ diff --git a/arch/riscv/kernel/Makefile b/arch/riscv/kernel/Makefile index 5b243d46f4b1..1e2afec141b5 100644 --- a/arch/riscv/kernel/Makefile +++ b/arch/riscv/kernel/Makefile @@ -110,3 +110,4 @@ obj-$(CONFIG_COMPAT) += compat_vdso/ obj-$(CONFIG_64BIT) += pi/ obj-$(CONFIG_ACPI) += acpi.o +obj-$(CONFIG_ACPI_NUMA) += acpi_numa.o diff --git a/arch/riscv/kernel/acpi.c b/arch/riscv/kernel/acpi.c index e619edc8b0cc..040bdbfea2b4 100644 --- a/arch/riscv/kernel/acpi.c +++ b/arch/riscv/kernel/acpi.c @@ -191,11 +191,6 @@ struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu) return &cpu_madt_rintc[cpu]; } -u32 get_acpi_id_for_cpu(int cpu) -{ - return acpi_cpu_get_madt_rintc(cpu)->uid; -} - /* * __acpi_map_table() will be called before paging_init(), so early_ioremap() * or early_memremap() should be called here to for ACPI table mapping. diff --git a/arch/riscv/kernel/acpi_numa.c b/arch/riscv/kernel/acpi_numa.c new file mode 100644 index 000000000000..0231482d6946 --- /dev/null +++ b/arch/riscv/kernel/acpi_numa.c @@ -0,0 +1,131 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * ACPI 6.6 based NUMA setup for RISCV + * Lots of code was borrowed from arch/arm64/kernel/acpi_numa.c + * + * Copyright 2004 Andi Kleen, SuSE Labs. + * Copyright (C) 2013-2016, Linaro Ltd. + * Author: Hanjun Guo + * Copyright (C) 2024 Intel Corporation. + * + * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs. + * + * Called from acpi_numa_init while reading the SRAT and SLIT tables. + * Assumes all memory regions belonging to a single proximity domain + * are in one chunk. Holes between them will be included in the node. + */ + +#define pr_fmt(fmt) "ACPI: NUMA: " fmt + +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +static int acpi_early_node_map[NR_CPUS] __initdata = { NUMA_NO_NODE }; + +int __init acpi_numa_get_nid(unsigned int cpu) +{ + return acpi_early_node_map[cpu]; +} + +static inline int get_cpu_for_acpi_id(u32 uid) +{ + int cpu; + + for (cpu = 0; cpu < nr_cpu_ids; cpu++) + if (uid == get_acpi_id_for_cpu(cpu)) + return cpu; + + return -EINVAL; +} + +static int __init acpi_parse_rintc_pxm(union acpi_subtable_headers *header, + const unsigned long end) +{ + struct acpi_srat_rintc_affinity *pa; + int cpu, pxm, node; + + if (srat_disabled()) + return -EINVAL; + + pa = (struct acpi_srat_rintc_affinity *)header; + if (!pa) + return -EINVAL; + + if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED)) + return 0; + + pxm = pa->proximity_domain; + node = pxm_to_node(pxm); + + /* + * If we can't map the UID to a logical cpu this + * means that the UID is not part of possible cpus + * so we do not need a NUMA mapping for it, skip + * the SRAT entry and keep parsing. + */ + cpu = get_cpu_for_acpi_id(pa->acpi_processor_uid); + if (cpu < 0) + return 0; + + acpi_early_node_map[cpu] = node; + pr_info("SRAT: PXM %d -> HARTID 0x%lx -> Node %d\n", pxm, + cpuid_to_hartid_map(cpu), node); + + return 0; +} + +void __init acpi_map_cpus_to_nodes(void) +{ + int i; + + /* + * In ACPI, SMP and CPU NUMA information is provided in separate + * static tables, namely the MADT and the SRAT. + * + * Thus, it is simpler to first create the cpu logical map through + * an MADT walk and then map the logical cpus to their node ids + * as separate steps. + */ + acpi_table_parse_entries(ACPI_SIG_SRAT, sizeof(struct acpi_table_srat), + ACPI_SRAT_TYPE_RINTC_AFFINITY, acpi_parse_rintc_pxm, 0); + + for (i = 0; i < nr_cpu_ids; i++) + early_map_cpu_to_node(i, acpi_numa_get_nid(i)); +} + +/* Callback for Proximity Domain -> logical node ID mapping */ +void __init acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) +{ + int pxm, node; + + if (srat_disabled()) + return; + + if (pa->header.length < sizeof(struct acpi_srat_rintc_affinity)) { + pr_err("SRAT: Invalid SRAT header length: %d\n", pa->header.length); + bad_srat(); + return; + } + + if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED)) + return; + + pxm = pa->proximity_domain; + node = acpi_map_pxm_to_node(pxm); + + if (node == NUMA_NO_NODE) { + pr_err("SRAT: Too many proximity domains %d\n", pxm); + bad_srat(); + return; + } + + node_set(node, numa_nodes_parsed); +} diff --git a/arch/riscv/kernel/setup.c b/arch/riscv/kernel/setup.c index 4f73c0ae44b2..a2cde65b69e9 100644 --- a/arch/riscv/kernel/setup.c +++ b/arch/riscv/kernel/setup.c @@ -281,8 +281,10 @@ void __init setup_arch(char **cmdline_p) setup_smp(); #endif - if (!acpi_disabled) + if (!acpi_disabled) { acpi_init_rintc_map(); + acpi_map_cpus_to_nodes(); + } riscv_init_cbo_blocksizes(); riscv_fill_hwcap(); diff --git a/arch/riscv/kernel/smpboot.c b/arch/riscv/kernel/smpboot.c index 1319b29ce3b5..d49c0bf0e2cd 100644 --- a/arch/riscv/kernel/smpboot.c +++ b/arch/riscv/kernel/smpboot.c @@ -96,7 +96,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un if (hart == cpuid_to_hartid_map(0)) { BUG_ON(found_boot_cpu); found_boot_cpu = true; - early_map_cpu_to_node(0, acpi_numa_get_nid(cpu_count)); return 0; } @@ -106,7 +105,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un } cpuid_to_hartid_map(cpu_count) = hart; - early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count)); cpu_count++; return 0; diff --git a/include/linux/acpi.h b/include/linux/acpi.h index 28c3fb2bef0d..5a2b620ccd58 100644 --- a/include/linux/acpi.h +++ b/include/linux/acpi.h @@ -264,6 +264,12 @@ static inline void acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { } #endif +#ifdef CONFIG_RISCV +void acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa); +#else +static inline void acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { } +#endif + #ifndef PHYS_CPUID_INVALID typedef u32 phys_cpuid_t; #define PHYS_CPUID_INVALID (phys_cpuid_t)(-1) -- cgit From 39494aec8a0454f9e8a68d2422d3bd4e5c5ee0c7 Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Thu, 13 Jun 2024 16:54:34 +0800 Subject: ACPI: NUMA: Add handler for SRAT RINTC affinity structure Add RINTC affinity structure handler during parsing SRAT table. Signed-off-by: Haibo Xu Reviewed-by: Sunil V L Reviewed-by: Hanjun Guo Link: https://lore.kernel.org/r/e076514d78d92f104a5f2d8c82b8921f6aa26fdd.1718268003.git.haibo1.xu@intel.com Signed-off-by: Palmer Dabbelt --- drivers/acpi/numa/srat.c | 32 +++++++++++++++++++++++++++++++- 1 file changed, 31 insertions(+), 1 deletion(-) diff --git a/drivers/acpi/numa/srat.c b/drivers/acpi/numa/srat.c index e3f26e71637a..44f91f2c6c5d 100644 --- a/drivers/acpi/numa/srat.c +++ b/drivers/acpi/numa/srat.c @@ -167,6 +167,19 @@ acpi_table_print_srat_entry(struct acpi_subtable_header *header) } } break; + + case ACPI_SRAT_TYPE_RINTC_AFFINITY: + { + struct acpi_srat_rintc_affinity *p = + (struct acpi_srat_rintc_affinity *)header; + pr_debug("SRAT Processor (acpi id[0x%04x]) in proximity domain %d %s\n", + p->acpi_processor_uid, + p->proximity_domain, + (p->flags & ACPI_SRAT_RINTC_ENABLED) ? + "enabled" : "disabled"); + } + break; + default: pr_warn("Found unsupported SRAT entry (type = 0x%x)\n", header->type); @@ -450,6 +463,21 @@ acpi_parse_gi_affinity(union acpi_subtable_headers *header, } #endif /* defined(CONFIG_X86) || defined (CONFIG_ARM64) */ +static int __init +acpi_parse_rintc_affinity(union acpi_subtable_headers *header, + const unsigned long end) +{ + struct acpi_srat_rintc_affinity *rintc_affinity; + + rintc_affinity = (struct acpi_srat_rintc_affinity *)header; + acpi_table_print_srat_entry(&header->common); + + /* let architecture-dependent part to do it */ + acpi_numa_rintc_affinity_init(rintc_affinity); + + return 0; +} + static int __init acpi_parse_srat(struct acpi_table_header *table) { struct acpi_table_srat *srat = (struct acpi_table_srat *)table; @@ -485,7 +513,7 @@ int __init acpi_numa_init(void) /* SRAT: System Resource Affinity Table */ if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) { - struct acpi_subtable_proc srat_proc[4]; + struct acpi_subtable_proc srat_proc[5]; memset(srat_proc, 0, sizeof(srat_proc)); srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY; @@ -496,6 +524,8 @@ int __init acpi_numa_init(void) srat_proc[2].handler = acpi_parse_gicc_affinity; srat_proc[3].id = ACPI_SRAT_TYPE_GENERIC_AFFINITY; srat_proc[3].handler = acpi_parse_gi_affinity; + srat_proc[4].id = ACPI_SRAT_TYPE_RINTC_AFFINITY; + srat_proc[4].handler = acpi_parse_rintc_affinity; acpi_table_parse_entries_array(ACPI_SIG_SRAT, sizeof(struct acpi_table_srat), -- cgit From adc3e82d253786a5b63c821108709de23317a90d Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Thu, 13 Jun 2024 16:54:35 +0800 Subject: ACPI: NUMA: change the ACPI_NUMA to a hidden option x86/arm64/loongarch would select ACPI_NUMA by default and riscv would do the same thing, so change it to a hidden option and the select statements except for the X86_64_ACPI_NUMA can also go away. Suggested-by: Arnd Bergmann Suggested-by: Sunil V L Signed-off-by: Haibo Xu Reviewed-by: Sunil V L Acked-by: Huacai Chen Acked-by: Will Deacon Reviewed-by: Hanjun Guo Link: https://lore.kernel.org/r/f1f96377b8ecd6e3183f28abf5c9ac21cb9855ea.1718268003.git.haibo1.xu@intel.com Signed-off-by: Palmer Dabbelt --- arch/arm64/Kconfig | 1 - arch/loongarch/Kconfig | 1 - drivers/acpi/numa/Kconfig | 5 +---- 3 files changed, 1 insertion(+), 6 deletions(-) diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index 5d91259ee7b5..5079ad4e21a5 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -1484,7 +1484,6 @@ config HOTPLUG_CPU config NUMA bool "NUMA Memory Allocation and Scheduler Support" select GENERIC_ARCH_NUMA - select ACPI_NUMA if ACPI select OF_NUMA select HAVE_SETUP_PER_CPU_AREA select NEED_PER_CPU_EMBED_FIRST_CHUNK diff --git a/arch/loongarch/Kconfig b/arch/loongarch/Kconfig index e38139c576ee..8d9e06e4ad84 100644 --- a/arch/loongarch/Kconfig +++ b/arch/loongarch/Kconfig @@ -470,7 +470,6 @@ config NR_CPUS config NUMA bool "NUMA Support" select SMP - select ACPI_NUMA if ACPI help Say Y to compile the kernel with NUMA (Non-Uniform Memory Access) support. This option improves performance on systems with more diff --git a/drivers/acpi/numa/Kconfig b/drivers/acpi/numa/Kconfig index 849c2bd820b9..f33194d1e43f 100644 --- a/drivers/acpi/numa/Kconfig +++ b/drivers/acpi/numa/Kconfig @@ -1,9 +1,6 @@ # SPDX-License-Identifier: GPL-2.0 config ACPI_NUMA - bool "NUMA support" - depends on NUMA - depends on (X86 || ARM64 || LOONGARCH) - default y if ARM64 + def_bool NUMA && !X86 config ACPI_HMAT bool "ACPI Heterogeneous Memory Attribute Table Support" -- cgit From 5f76d4211ee44dcd9b2dc8a9065002bf8ecee81e Mon Sep 17 00:00:00 2001 From: Haibo Xu Date: Thu, 13 Jun 2024 16:54:36 +0800 Subject: ACPI: NUMA: replace pr_info with pr_debug in arch_acpi_numa_init There are lots of ACPI enabled systems that aren't NUMA and If the firmware didn't provide the SRAT/SLIT, then there will be a message "Failed to initialise from firmware" from arch_acpi_numa_init() which adding noise to the boot on all of those kind of systems. Replace the pr_info with pr_debug in arch_acpi_numa_init() to avoid it. Suggested-by: Sunil V L Signed-off-by: Haibo Xu Reviewed-by: Sunil V L Reviewed-by: Hanjun Guo Link: https://lore.kernel.org/r/109354315a02cd22145d2effa4a8c571b69d3e56.1718268003.git.haibo1.xu@intel.com Signed-off-by: Palmer Dabbelt --- drivers/base/arch_numa.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/base/arch_numa.c b/drivers/base/arch_numa.c index 5b59d133b6af..555aee3ee8e7 100644 --- a/drivers/base/arch_numa.c +++ b/drivers/base/arch_numa.c @@ -445,7 +445,7 @@ static int __init arch_acpi_numa_init(void) ret = acpi_numa_init(); if (ret) { - pr_info("Failed to initialise from firmware\n"); + pr_debug("Failed to initialise from firmware\n"); return ret; } -- cgit