LoongArch: Add Non-Uniform Memory Access (NUMA) support

Add Non-Uniform Memory Access (NUMA) support for LoongArch. LoongArch
has 48-bit physical address, but the HyperTransport I/O bus only support
40-bit address, so we need a custom phys_to_dma() and dma_to_phys() to
extract the 4-bit node id (bit 44~47) from Loongson-3's 48-bit physical
address space and embed it into 40-bit. In the 40-bit dma address, node
id offset can be read from the LS7A_DMA_CFG register.

Reviewed-by: WANG Xuerui <git@xen0n.name>
Reviewed-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>

8 files changed, 649 insertions, 22 deletions

diff --git a/arch/loongarch/kernel/Makefile b/arch/loongarch/kernel/Makefile
index 2cb6f698716a..940de9173542 100644
--- a/arch/loongarch/kernel/Makefile
+++ b/arch/loongarch/kernel/Makefile
@@ -20,4 +20,6 @@ obj-$(CONFIG_PROC_FS)		+= proc.o
 
 obj-$(CONFIG_SMP)		+= smp.o
 
+obj-$(CONFIG_NUMA)		+= numa.o
+
 CPPFLAGS_vmlinux.lds		:= $(KBUILD_CFLAGS)
diff --git a/arch/loongarch/kernel/acpi.c b/arch/loongarch/kernel/acpi.c
index 181c59493b63..b16c3dea5eeb 100644
--- a/arch/loongarch/kernel/acpi.c
+++ b/arch/loongarch/kernel/acpi.c
@@ -14,6 +14,7 @@
 #include <linux/memblock.h>
 #include <linux/serial_core.h>
 #include <asm/io.h>
+#include <asm/numa.h>
 #include <asm/loongson.h>
 
 int acpi_disabled;
@@ -199,6 +200,79 @@ int __init acpi_boot_init(void)
 	return 0;
 }
 
+#ifdef CONFIG_ACPI_NUMA
+
+static __init int setup_node(int pxm)
+{
+	return acpi_map_pxm_to_node(pxm);
+}
+
+/*
+ * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
+ * I/O localities since SRAT does not list them.  I/O localities are
+ * not supported at this point.
+ */
+unsigned int numa_distance_cnt;
+
+static inline unsigned int get_numa_distances_cnt(struct acpi_table_slit *slit)
+{
+	return slit->locality_count;
+}
+
+void __init numa_set_distance(int from, int to, int distance)
+{
+	if ((u8)distance != distance || (from == to && distance != LOCAL_DISTANCE)) {
+		pr_warn_once("Warning: invalid distance parameter, from=%d to=%d distance=%d\n",
+				from, to, distance);
+		return;
+	}
+
+	node_distances[from][to] = distance;
+}
+
+/* Callback for Proximity Domain -> CPUID mapping */
+void __init
+acpi_numa_processor_affinity_init(struct acpi_srat_cpu_affinity *pa)
+{
+	int pxm, node;
+
+	if (srat_disabled())
+		return;
+	if (pa->header.length != sizeof(struct acpi_srat_cpu_affinity)) {
+		bad_srat();
+		return;
+	}
+	if ((pa->flags & ACPI_SRAT_CPU_ENABLED) == 0)
+		return;
+	pxm = pa->proximity_domain_lo;
+	if (acpi_srat_revision >= 2) {
+		pxm |= (pa->proximity_domain_hi[0] << 8);
+		pxm |= (pa->proximity_domain_hi[1] << 16);
+		pxm |= (pa->proximity_domain_hi[2] << 24);
+	}
+	node = setup_node(pxm);
+	if (node < 0) {
+		pr_err("SRAT: Too many proximity domains %x\n", pxm);
+		bad_srat();
+		return;
+	}
+
+	if (pa->apic_id >= CONFIG_NR_CPUS) {
+		pr_info("SRAT: PXM %u -> CPU 0x%02x -> Node %u skipped apicid that is too big\n",
+				pxm, pa->apic_id, node);
+		return;
+	}
+
+	early_numa_add_cpu(pa->apic_id, node);
+
+	set_cpuid_to_node(pa->apic_id, node);
+	node_set(node, numa_nodes_parsed);
+	pr_info("SRAT: PXM %u -> CPU 0x%02x -> Node %u\n", pxm, pa->apic_id, node);
+}
+
+void __init acpi_numa_arch_fixup(void) {}
+#endif
+
 void __init arch_reserve_mem_area(acpi_physical_address addr, size_t size)
 {
 	memblock_reserve(addr, size);
@@ -208,6 +282,22 @@ void __init arch_reserve_mem_area(acpi_physical_address addr, size_t size)
 
 #include <acpi/processor.h>
 
+static int __ref acpi_map_cpu2node(acpi_handle handle, int cpu, int physid)
+{
+#ifdef CONFIG_ACPI_NUMA
+	int nid;
+
+	nid = acpi_get_node(handle);
+	if (nid != NUMA_NO_NODE) {
+		set_cpuid_to_node(physid, nid);
+		node_set(nid, numa_nodes_parsed);
+		set_cpu_numa_node(cpu, nid);
+		cpumask_set_cpu(cpu, cpumask_of_node(nid));
+	}
+#endif
+	return 0;
+}
+
 int acpi_map_cpu(acpi_handle handle, phys_cpuid_t physid, u32 acpi_id, int *pcpu)
 {
 	int cpu;
@@ -218,6 +308,8 @@ int acpi_map_cpu(acpi_handle handle, phys_cpuid_t physid, u32 acpi_id, int *pcpu
 		return cpu;
 	}
 
+	acpi_map_cpu2node(handle, cpu, physid);
+
 	*pcpu = cpu;
 
 	return 0;
@@ -226,6 +318,9 @@ EXPORT_SYMBOL(acpi_map_cpu);
 
 int acpi_unmap_cpu(int cpu)
 {
+#ifdef CONFIG_ACPI_NUMA
+	set_cpuid_to_node(cpu_logical_map(cpu), NUMA_NO_NODE);
+#endif
 	set_cpu_present(cpu, false);
 	num_processors--;
 
diff --git a/arch/loongarch/kernel/dma.c b/arch/loongarch/kernel/dma.c
new file mode 100644
index 000000000000..8c9b5314a13e
--- /dev/null
+++ b/arch/loongarch/kernel/dma.c
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#include <linux/init.h>
+#include <linux/dma-direct.h>
+#include <linux/dma-mapping.h>
+#include <linux/dma-map-ops.h>
+#include <linux/swiotlb.h>
+
+#include <asm/bootinfo.h>
+#include <asm/dma.h>
+#include <asm/loongson.h>
+
+/*
+ * We extract 4bit node id (bit 44~47) from Loongson-3's
+ * 48bit physical address space and embed it into 40bit.
+ */
+
+static int node_id_offset;
+
+dma_addr_t phys_to_dma(struct device *dev, phys_addr_t paddr)
+{
+	long nid = (paddr >> 44) & 0xf;
+
+	return ((nid << 44) ^ paddr) | (nid << node_id_offset);
+}
+
+phys_addr_t dma_to_phys(struct device *dev, dma_addr_t daddr)
+{
+	long nid = (daddr >> node_id_offset) & 0xf;
+
+	return ((nid << node_id_offset) ^ daddr) | (nid << 44);
+}
+
+void __init plat_swiotlb_setup(void)
+{
+	swiotlb_init(true, SWIOTLB_VERBOSE);
+	node_id_offset = ((readl(LS7A_DMA_CFG) & LS7A_DMA_NODE_MASK) >> LS7A_DMA_NODE_SHF) + 36;
+}
diff --git a/arch/loongarch/kernel/module.c b/arch/loongarch/kernel/module.c
index 87b3768f1eef..638427ff0d51 100644
--- a/arch/loongarch/kernel/module.c
+++ b/arch/loongarch/kernel/module.c
@@ -11,6 +11,7 @@
 #include <linux/moduleloader.h>
 #include <linux/elf.h>
 #include <linux/mm.h>
+#include <linux/numa.h>
 #include <linux/vmalloc.h>
 #include <linux/slab.h>
 #include <linux/fs.h>
diff --git a/arch/loongarch/kernel/numa.c b/arch/loongarch/kernel/numa.c
new file mode 100644
index 000000000000..a76f547a5aa3
--- /dev/null
+++ b/arch/loongarch/kernel/numa.c
@@ -0,0 +1,466 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Author:  Xiang Gao <gaoxiang@loongson.cn>
+ *          Huacai Chen <chenhuacai@loongson.cn>
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/mmzone.h>
+#include <linux/export.h>
+#include <linux/nodemask.h>
+#include <linux/swap.h>
+#include <linux/memblock.h>
+#include <linux/pfn.h>
+#include <linux/acpi.h>
+#include <linux/efi.h>
+#include <linux/irq.h>
+#include <linux/pci.h>
+#include <asm/bootinfo.h>
+#include <asm/loongson.h>
+#include <asm/numa.h>
+#include <asm/page.h>
+#include <asm/pgalloc.h>
+#include <asm/sections.h>
+#include <asm/time.h>
+
+int numa_off;
+struct pglist_data *node_data[MAX_NUMNODES];
+unsigned char node_distances[MAX_NUMNODES][MAX_NUMNODES];
+
+EXPORT_SYMBOL(node_data);
+EXPORT_SYMBOL(node_distances);
+
+static struct numa_meminfo numa_meminfo;
+cpumask_t cpus_on_node[MAX_NUMNODES];
+cpumask_t phys_cpus_on_node[MAX_NUMNODES];
+EXPORT_SYMBOL(cpus_on_node);
+
+/*
+ * apicid, cpu, node mappings
+ */
+s16 __cpuid_to_node[CONFIG_NR_CPUS] = {
+	[0 ... CONFIG_NR_CPUS - 1] = NUMA_NO_NODE
+};
+EXPORT_SYMBOL(__cpuid_to_node);
+
+nodemask_t numa_nodes_parsed __initdata;
+
+#ifdef CONFIG_HAVE_SETUP_PER_CPU_AREA
+unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
+EXPORT_SYMBOL(__per_cpu_offset);
+
+static int __init pcpu_cpu_to_node(int cpu)
+{
+	return early_cpu_to_node(cpu);
+}
+
+static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
+{
+	if (early_cpu_to_node(from) == early_cpu_to_node(to))
+		return LOCAL_DISTANCE;
+	else
+		return REMOTE_DISTANCE;
+}
+
+void __init pcpu_populate_pte(unsigned long addr)
+{
+	pgd_t *pgd = pgd_offset_k(addr);
+	p4d_t *p4d = p4d_offset(pgd, addr);
+	pud_t *pud;
+	pmd_t *pmd;
+
+	if (p4d_none(*p4d)) {
+		pud_t *new;
+
+		new = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
+		pgd_populate(&init_mm, pgd, new);
+#ifndef __PAGETABLE_PUD_FOLDED
+		pud_init((unsigned long)new, (unsigned long)invalid_pmd_table);
+#endif
+	}
+
+	pud = pud_offset(p4d, addr);
+	if (pud_none(*pud)) {
+		pmd_t *new;
+
+		new = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
+		pud_populate(&init_mm, pud, new);
+#ifndef __PAGETABLE_PMD_FOLDED
+		pmd_init((unsigned long)new, (unsigned long)invalid_pte_table);
+#endif
+	}
+
+	pmd = pmd_offset(pud, addr);
+	if (!pmd_present(*pmd)) {
+		pte_t *new;
+
+		new = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
+		pmd_populate_kernel(&init_mm, pmd, new);
+	}
+}
+
+void __init setup_per_cpu_areas(void)
+{
+	unsigned long delta;
+	unsigned int cpu;
+	int rc = -EINVAL;
+
+	if (pcpu_chosen_fc == PCPU_FC_AUTO) {
+		if (nr_node_ids >= 8)
+			pcpu_chosen_fc = PCPU_FC_PAGE;
+		else
+			pcpu_chosen_fc = PCPU_FC_EMBED;
+	}
+
+	/*
+	 * Always reserve area for module percpu variables.  That's
+	 * what the legacy allocator did.
+	 */
+	if (pcpu_chosen_fc != PCPU_FC_PAGE) {
+		rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
+					    PERCPU_DYNAMIC_RESERVE, PMD_SIZE,
+					    pcpu_cpu_distance, pcpu_cpu_to_node);
+		if (rc < 0)
+			pr_warn("%s allocator failed (%d), falling back to page size\n",
+				pcpu_fc_names[pcpu_chosen_fc], rc);
+	}
+	if (rc < 0)
+		rc = pcpu_page_first_chunk(PERCPU_MODULE_RESERVE, pcpu_cpu_to_node);
+	if (rc < 0)
+		panic("cannot initialize percpu area (err=%d)", rc);
+
+	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
+	for_each_possible_cpu(cpu)
+		__per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
+}
+#endif
+
+/*
+ * Get nodeid by logical cpu number.
+ * __cpuid_to_node maps phyical cpu id to node, so we
+ * should use cpu_logical_map(cpu) to index it.
+ *
+ * This routine is only used in early phase during
+ * booting, after setup_per_cpu_areas calling and numa_node
+ * initialization, cpu_to_node will be used instead.
+ */
+int early_cpu_to_node(int cpu)
+{
+	int physid = cpu_logical_map(cpu);
+
+	if (physid < 0)
+		return NUMA_NO_NODE;
+
+	return __cpuid_to_node[physid];
+}
+
+void __init early_numa_add_cpu(int cpuid, s16 node)
+{
+	int cpu = __cpu_number_map[cpuid];
+
+	if (cpu < 0)
+		return;
+
+	cpumask_set_cpu(cpu, &cpus_on_node[node]);
+	cpumask_set_cpu(cpuid, &phys_cpus_on_node[node]);
+}
+
+void numa_add_cpu(unsigned int cpu)
+{
+	int nid = cpu_to_node(cpu);
+	cpumask_set_cpu(cpu, &cpus_on_node[nid]);
+}
+
+void numa_remove_cpu(unsigned int cpu)
+{
+	int nid = cpu_to_node(cpu);
+	cpumask_clear_cpu(cpu, &cpus_on_node[nid]);
+}
+
+static int __init numa_add_memblk_to(int nid, u64 start, u64 end,
+				     struct numa_meminfo *mi)
+{
+	/* ignore zero length blks */
+	if (start == end)
+		return 0;
+
+	/* whine about and ignore invalid blks */
+	if (start > end || nid < 0 || nid >= MAX_NUMNODES) {
+		pr_warn("NUMA: Warning: invalid memblk node %d [mem %#010Lx-%#010Lx]\n",
+			   nid, start, end - 1);
+		return 0;
+	}
+
+	if (mi->nr_blks >= NR_NODE_MEMBLKS) {
+		pr_err("NUMA: too many memblk ranges\n");
+		return -EINVAL;
+	}
+
+	mi->blk[mi->nr_blks].start = PFN_ALIGN(start);
+	mi->blk[mi->nr_blks].end = PFN_ALIGN(end - PAGE_SIZE + 1);
+	mi->blk[mi->nr_blks].nid = nid;
+	mi->nr_blks++;
+	return 0;
+}
+
+/**
+ * numa_add_memblk - Add one numa_memblk to numa_meminfo
+ * @nid: NUMA node ID of the new memblk
+ * @start: Start address of the new memblk
+ * @end: End address of the new memblk
+ *
+ * Add a new memblk to the default numa_meminfo.
+ *
+ * RETURNS:
+ * 0 on success, -errno on failure.
+ */
+int __init numa_add_memblk(int nid, u64 start, u64 end)
+{
+	return numa_add_memblk_to(nid, start, end, &numa_meminfo);
+}
+
+static void __init alloc_node_data(int nid)
+{
+	void *nd;
+	unsigned long nd_pa;
+	size_t nd_sz = roundup(sizeof(pg_data_t), PAGE_SIZE);
+
+	nd_pa = memblock_phys_alloc_try_nid(nd_sz, SMP_CACHE_BYTES, nid);
+	if (!nd_pa) {
+		pr_err("Cannot find %zu Byte for node_data (initial node: %d)\n", nd_sz, nid);
+		return;
+	}
+
+	nd = __va(nd_pa);
+
+	node_data[nid] = nd;
+	memset(nd, 0, sizeof(pg_data_t));
+}
+
+static void __init node_mem_init(unsigned int node)
+{
+	unsigned long start_pfn, end_pfn;
+	unsigned long node_addrspace_offset;
+
+	node_addrspace_offset = nid_to_addrbase(node);
+	pr_info("Node%d's addrspace_offset is 0x%lx\n",
+			node, node_addrspace_offset);
+
+	get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
+	pr_info("Node%d: start_pfn=0x%lx, end_pfn=0x%lx\n",
+		node, start_pfn, end_pfn);
+
+	alloc_node_data(node);
+}
+
+#ifdef CONFIG_ACPI_NUMA
+
+/*
+ * Sanity check to catch more bad NUMA configurations (they are amazingly
+ * common).  Make sure the nodes cover all memory.
+ */
+static bool __init numa_meminfo_cover_memory(const struct numa_meminfo *mi)
+{
+	int i;
+	u64 numaram, biosram;
+
+	numaram = 0;
+	for (i = 0; i < mi->nr_blks; i++) {
+		u64 s = mi->blk[i].start >> PAGE_SHIFT;
+		u64 e = mi->blk[i].end >> PAGE_SHIFT;
+
+		numaram += e - s;
+		numaram -= __absent_pages_in_range(mi->blk[i].nid, s, e);
+		if ((s64)numaram < 0)
+			numaram = 0;
+	}
+	max_pfn = max_low_pfn;
+	biosram = max_pfn - absent_pages_in_range(0, max_pfn);
+
+	BUG_ON((s64)(biosram - numaram) >= (1 << (20 - PAGE_SHIFT)));
+	return true;
+}
+
+static void __init add_node_intersection(u32 node, u64 start, u64 size, u32 type)
+{
+	static unsigned long num_physpages;
+
+	num_physpages += (size >> PAGE_SHIFT);
+	pr_info("Node%d: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx Bytes\n",
+		node, type, start, size);
+	pr_info("       start_pfn:0x%llx, end_pfn:0x%llx, num_physpages:0x%lx\n",
+		start >> PAGE_SHIFT, (start + size) >> PAGE_SHIFT, num_physpages);
+	memblock_set_node(start, size, &memblock.memory, node);
+}
+
+/*
+ * add_numamem_region
+ *
+ * Add a uasable memory region described by BIOS. The
+ * routine gets each intersection between BIOS's region
+ * and node's region, and adds them into node's memblock
+ * pool.
+ *
+ */
+static void __init add_numamem_region(u64 start, u64 end, u32 type)
+{
+	u32 i;
+	u64 ofs = start;
+
+	if (start >= end) {
+		pr_debug("Invalid region: %016llx-%016llx\n", start, end);
+		return;
+	}
+
+	for (i = 0; i < numa_meminfo.nr_blks; i++) {
+		struct numa_memblk *mb = &numa_meminfo.blk[i];
+
+		if (ofs > mb->end)
+			continue;
+
+		if (end > mb->end) {
+			add_node_intersection(mb->nid, ofs, mb->end - ofs, type);
+			ofs = mb->end;
+		} else {
+			add_node_intersection(mb->nid, ofs, end - ofs, type);
+			break;
+		}
+	}
+}
+
+static void __init init_node_memblock(void)
+{
+	u32 mem_type;
+	u64 mem_end, mem_start, mem_size;
+	efi_memory_desc_t *md;
+
+	/* Parse memory information and activate */
+	for_each_efi_memory_desc(md) {
+		mem_type = md->type;
+		mem_start = md->phys_addr;
+		mem_size = md->num_pages << EFI_PAGE_SHIFT;
+		mem_end = mem_start + mem_size;
+
+		switch (mem_type) {
+		case EFI_LOADER_CODE:
+		case EFI_LOADER_DATA:
+		case EFI_BOOT_SERVICES_CODE:
+		case EFI_BOOT_SERVICES_DATA:
+		case EFI_PERSISTENT_MEMORY:
+		case EFI_CONVENTIONAL_MEMORY:
+			add_numamem_region(mem_start, mem_end, mem_type);
+			break;
+		case EFI_PAL_CODE:
+		case EFI_UNUSABLE_MEMORY:
+		case EFI_ACPI_RECLAIM_MEMORY:
+			add_numamem_region(mem_start, mem_end, mem_type);
+			fallthrough;
+		case EFI_RESERVED_TYPE:
+		case EFI_RUNTIME_SERVICES_CODE:
+		case EFI_RUNTIME_SERVICES_DATA:
+		case EFI_MEMORY_MAPPED_IO:
+		case EFI_MEMORY_MAPPED_IO_PORT_SPACE:
+			pr_info("Resvd: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx Bytes\n",
+					mem_type, mem_start, mem_size);
+			break;
+		}
+	}
+}
+
+static void __init numa_default_distance(void)
+{
+	int row, col;
+
+	for (row = 0; row < MAX_NUMNODES; row++)
+		for (col = 0; col < MAX_NUMNODES; col++) {
+			if (col == row)
+				node_distances[row][col] = LOCAL_DISTANCE;
+			else
+				/* We assume that one node per package here!
+				 *
+				 * A SLIT should be used for multiple nodes
+				 * per package to override default setting.
+				 */
+				node_distances[row][col] = REMOTE_DISTANCE;
+	}
+}
+
+int __init init_numa_memory(void)
+{
+	int i;
+	int ret;
+	int node;
+
+	for (i = 0; i < NR_CPUS; i++)
+		set_cpuid_to_node(i, NUMA_NO_NODE);
+
+	numa_default_distance();
+	nodes_clear(numa_nodes_parsed);
+	nodes_clear(node_possible_map);
+	nodes_clear(node_online_map);
+	memset(&numa_meminfo, 0, sizeof(numa_meminfo));
+
+	/* Parse SRAT and SLIT if provided by firmware. */
+	ret = acpi_numa_init();
+	if (ret < 0)
+		return ret;
+
+	node_possible_map = numa_nodes_parsed;
+	if (WARN_ON(nodes_empty(node_possible_map)))
+		return -EINVAL;
+
+	init_node_memblock();
+	if (numa_meminfo_cover_memory(&numa_meminfo) == false)
+		return -EINVAL;
+
+	for_each_node_mask(node, node_possible_map) {
+		node_mem_init(node);
+		node_set_online(node);
+	}
+	max_low_pfn = PHYS_PFN(memblock_end_of_DRAM());
+
+	setup_nr_node_ids();
+	loongson_sysconf.nr_nodes = nr_node_ids;
+	loongson_sysconf.cores_per_node = cpumask_weight(&phys_cpus_on_node[0]);
+
+	return 0;
+}
+
+EXPORT_SYMBOL(init_numa_memory);
+#endif
+
+void __init paging_init(void)
+{
+	unsigned int node;
+	unsigned long zones_size[MAX_NR_ZONES] = {0, };
+
+	for_each_online_node(node) {
+		unsigned long start_pfn, end_pfn;
+
+		get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
+
+		if (end_pfn > max_low_pfn)
+			max_low_pfn = end_pfn;
+	}
+#ifdef CONFIG_ZONE_DMA32
+	zones_size[ZONE_DMA32] = MAX_DMA32_PFN;
+#endif
+	zones_size[ZONE_NORMAL] = max_low_pfn;
+	free_area_init(zones_size);
+}
+
+void __init mem_init(void)
+{
+	high_memory = (void *) __va(get_num_physpages() << PAGE_SHIFT);
+	memblock_free_all();
+	setup_zero_pages();	/* This comes from node 0 */
+}
+
+int pcibus_to_node(struct pci_bus *bus)
+{
+	return dev_to_node(&bus->dev);
+}
+EXPORT_SYMBOL(pcibus_to_node);
diff --git a/arch/loongarch/kernel/setup.c b/arch/loongarch/kernel/setup.c
index 34a3011f679e..185e4035811a 100644
--- a/arch/loongarch/kernel/setup.c
+++ b/arch/loongarch/kernel/setup.c
@@ -35,6 +35,7 @@
 #include <asm/dma.h>
 #include <asm/efi.h>
 #include <asm/loongson.h>
+#include <asm/numa.h>
 #include <asm/pgalloc.h>
 #include <asm/sections.h>
 #include <asm/setup.h>
@@ -185,7 +186,10 @@ static int __init early_parse_mem(char *p)
 		return -EINVAL;
 	}
 
-	memblock_add(start, size);
+	if (!IS_ENABLED(CONFIG_NUMA))
+		memblock_add(start, size);
+	else
+		memblock_add_node(start, size, pa_to_nid(start), MEMBLOCK_NONE);
 
 	return 0;
 }
@@ -203,6 +207,9 @@ void __init platform_init(void)
 	acpi_boot_init();
 #endif
 
+#ifdef CONFIG_NUMA
+	init_numa_memory();
+#endif
 	dmi_setup();
 	smbios_parse();
 	pr_info("The BIOS Version: %s\n", b_info.bios_version);
@@ -241,7 +248,7 @@ static void __init arch_mem_init(char **cmdline_p)
 	sparse_init();
 	memblock_set_bottom_up(true);
 
-	swiotlb_init(true, SWIOTLB_VERBOSE);
+	plat_swiotlb_setup();
 
 	dma_contiguous_reserve(PFN_PHYS(max_low_pfn));
 
diff --git a/arch/loongarch/kernel/smp.c b/arch/loongarch/kernel/smp.c
index 99ba7a56edf9..b8c53b755a25 100644
--- a/arch/loongarch/kernel/smp.c
+++ b/arch/loongarch/kernel/smp.c
@@ -25,6 +25,7 @@
 #include <asm/idle.h>
 #include <asm/loongson.h>
 #include <asm/mmu_context.h>
+#include <asm/numa.h>
 #include <asm/processor.h>
 #include <asm/setup.h>
 #include <asm/time.h>
@@ -222,6 +223,9 @@ void loongson3_init_secondary(void)
 
 	iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
 
+#ifdef CONFIG_NUMA
+	numa_add_cpu(cpu);
+#endif
 	per_cpu(cpu_state, cpu) = CPU_ONLINE;
 	cpu_data[cpu].core =
 		     cpu_logical_map(cpu) % loongson_sysconf.cores_per_package;
@@ -254,6 +258,9 @@ int loongson3_cpu_disable(void)
 	if (io_master(cpu))
 		return -EBUSY;
 
+#ifdef CONFIG_NUMA
+	numa_remove_cpu(cpu);
+#endif
 	set_cpu_online(cpu, false);
 	calculate_cpu_foreign_map();
 	local_irq_save(flags);
@@ -478,14 +485,36 @@ void calculate_cpu_foreign_map(void)
 /* Preload SMP state for boot cpu */
 void smp_prepare_boot_cpu(void)
 {
-	unsigned int cpu;
+	unsigned int cpu, node, rr_node;
 
 	set_cpu_possible(0, true);
 	set_cpu_online(0, true);
 	set_my_cpu_offset(per_cpu_offset(0));
 
-	for_each_possible_cpu(cpu)
-		set_cpu_numa_node(cpu, 0);
+	rr_node = first_node(node_online_map);
+	for_each_possible_cpu(cpu) {
+		node = early_cpu_to_node(cpu);
+
+		/*
+		 * The mapping between present cpus and nodes has been
+		 * built during MADT and SRAT parsing.
+		 *
+		 * If possible cpus = present cpus here, early_cpu_to_node
+		 * will return valid node.
+		 *
+		 * If possible cpus > present cpus here (e.g. some possible
+		 * cpus will be added by cpu-hotplug later), for possible but
+		 * not present cpus, early_cpu_to_node will return NUMA_NO_NODE,
+		 * and we just map them to online nodes in round-robin way.
+		 * Once hotplugged, new correct mapping will be built for them.
+		 */
+		if (node != NUMA_NO_NODE)
+			set_cpu_numa_node(cpu, node);
+		else {
+			set_cpu_numa_node(cpu, rr_node);
+			rr_node = next_node_in(rr_node, node_online_map);
+		}
+	}
 }
 
 /* called from main before smp_init() */
@@ -651,17 +680,10 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned l
 		on_each_cpu_mask(mm_cpumask(mm), flush_tlb_range_ipi, &fd, 1);
 	} else {
 		unsigned int cpu;
-		int exec = vma->vm_flags & VM_EXEC;
 
 		for_each_online_cpu(cpu) {
-			/*
-			 * flush_cache_range() will only fully flush icache if
-			 * the VMA is executable, otherwise we must invalidate
-			 * ASID without it appearing to has_valid_asid() as if
-			 * mm has been completely unused by that CPU.
-			 */
 			if (cpu != smp_processor_id() && cpu_context(cpu, mm))
-				cpu_context(cpu, mm) = !exec;
+				cpu_context(cpu, mm) = 0;
 		}
 		local_flush_tlb_range(vma, start, end);
 	}
@@ -706,14 +728,8 @@ void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 		unsigned int cpu;
 
 		for_each_online_cpu(cpu) {
-			/*
-			 * flush_cache_page() only does partial flushes, so
-			 * invalidate ASID without it appearing to
-			 * has_valid_asid() as if mm has been completely unused
-			 * by that CPU.
-			 */
 			if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
-				cpu_context(cpu, vma->vm_mm) = 1;
+				cpu_context(cpu, vma->vm_mm) = 0;
 		}
 		local_flush_tlb_page(vma, page);
 	}
diff --git a/arch/loongarch/kernel/traps.c b/arch/loongarch/kernel/traps.c
index 8be071b265b3..e4060f84a221 100644
--- a/arch/loongarch/kernel/traps.c
+++ b/arch/loongarch/kernel/traps.c
@@ -630,7 +630,7 @@ asmlinkage void noinstr do_vint(struct pt_regs *regs, unsigned long sp)
 	irqentry_exit(regs, state);
 }
 
-extern void tlb_init(void);
+extern void tlb_init(int cpu);
 extern void cache_error_setup(void);
 
 unsigned long eentry;
@@ -669,7 +669,7 @@ void per_cpu_trap_init(int cpu)
 		for (i = 0; i < 64; i++)
 			set_handler(i * VECSIZE, handle_reserved, VECSIZE);
 
-	tlb_init();
+	tlb_init(cpu);
 	cpu_cache_init();
 }