1 files changed, 339 insertions, 28 deletions

diff --git a/arch/x86/kernel/setup_percpu.c b/arch/x86/kernel/setup_percpu.c
index d992e6cff730..c29f301d3885 100644
--- a/arch/x86/kernel/setup_percpu.c
+++ b/arch/x86/kernel/setup_percpu.c
@@ -7,6 +7,7 @@
 #include <linux/crash_dump.h>
 #include <linux/smp.h>
 #include <linux/topology.h>
+#include <linux/pfn.h>
 #include <asm/sections.h>
 #include <asm/processor.h>
 #include <asm/setup.h>
@@ -41,6 +42,321 @@ unsigned long __per_cpu_offset[NR_CPUS] __read_mostly = {
 };
 EXPORT_SYMBOL(__per_cpu_offset);
 
+/**
+ * pcpu_need_numa - determine percpu allocation needs to consider NUMA
+ *
+ * If NUMA is not configured or there is only one NUMA node available,
+ * there is no reason to consider NUMA.  This function determines
+ * whether percpu allocation should consider NUMA or not.
+ *
+ * RETURNS:
+ * true if NUMA should be considered; otherwise, false.
+ */
+static bool __init pcpu_need_numa(void)
+{
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+        pg_data_t *last = NULL;
+        unsigned int cpu;
+
+        for_each_possible_cpu(cpu) {
+                int node = early_cpu_to_node(cpu);
+
+                if (node_online(node) && NODE_DATA(node) &&
+                    last && last != NODE_DATA(node))
+                        return true;
+
+                last = NODE_DATA(node);
+        }
+#endif
+        return false;
+}
+
+/**
+ * pcpu_alloc_bootmem - NUMA friendly alloc_bootmem wrapper for percpu
+ * @cpu: cpu to allocate for
+ * @size: size allocation in bytes
+ * @align: alignment
+ *
+ * Allocate @size bytes aligned at @align for cpu @cpu.  This wrapper
+ * does the right thing for NUMA regardless of the current
+ * configuration.
+ *
+ * RETURNS:
+ * Pointer to the allocated area on success, NULL on failure.
+ */
+static void * __init pcpu_alloc_bootmem(unsigned int cpu, unsigned long size,
+                                        unsigned long align)
+{
+        const unsigned long goal = __pa(MAX_DMA_ADDRESS);
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+        int node = early_cpu_to_node(cpu);
+        void *ptr;
+
+        if (!node_online(node) || !NODE_DATA(node)) {
+                ptr = __alloc_bootmem_nopanic(size, align, goal);
+                pr_info("cpu %d has no node %d or node-local memory\n",
+                        cpu, node);
+                pr_debug("per cpu data for cpu%d %lu bytes at %016lx\n",
+                         cpu, size, __pa(ptr));
+        } else {
+                ptr = __alloc_bootmem_node_nopanic(NODE_DATA(node),
+                                                   size, align, goal);
+                pr_debug("per cpu data for cpu%d %lu bytes on node%d at "
+                         "%016lx\n", cpu, size, node, __pa(ptr));
+        }
+        return ptr;
+#else
+        return __alloc_bootmem_nopanic(size, align, goal);
+#endif
+}
+
+/*
+ * Remap allocator
+ *
+ * This allocator uses PMD page as unit.  A PMD page is allocated for
+ * each cpu and each is remapped into vmalloc area using PMD mapping.
+ * As PMD page is quite large, only part of it is used for the first
+ * chunk.  Unused part is returned to the bootmem allocator.
+ *
+ * So, the PMD pages are mapped twice - once to the physical mapping
+ * and to the vmalloc area for the first percpu chunk.  The double
+ * mapping does add one more PMD TLB entry pressure but still is much
+ * better than only using 4k mappings while still being NUMA friendly.
+ */
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+static size_t pcpur_size __initdata;
+static void **pcpur_ptrs __initdata;
+
+static struct page * __init pcpur_get_page(unsigned int cpu, int pageno)
+{
+        size_t off = (size_t)pageno << PAGE_SHIFT;
+
+        if (off >= pcpur_size)
+                return NULL;
+
+        return virt_to_page(pcpur_ptrs[cpu] + off);
+}
+
+static ssize_t __init setup_pcpu_remap(size_t static_size)
+{
+        static struct vm_struct vm;
+        pg_data_t *last;
+        size_t ptrs_size;
+        unsigned int cpu;
+        ssize_t ret;
+
+        /*
+         * If large page isn't supported, there's no benefit in doing
+         * this.  Also, on non-NUMA, embedding is better.
+         */
+        if (!cpu_has_pse || pcpu_need_numa())
+                return -EINVAL;
+
+        last = NULL;
+        for_each_possible_cpu(cpu) {
+                int node = early_cpu_to_node(cpu);
+
+                if (node_online(node) && NODE_DATA(node) &&
+                    last && last != NODE_DATA(node))
+                        goto proceed;
+
+                last = NODE_DATA(node);
+        }
+        return -EINVAL;
+
+proceed:
+        /*
+         * Currently supports only single page.  Supporting multiple
+         * pages won't be too difficult if it ever becomes necessary.
+         */
+        pcpur_size = PFN_ALIGN(static_size + PERCPU_DYNAMIC_RESERVE);
+        if (pcpur_size > PMD_SIZE) {
+                pr_warning("PERCPU: static data is larger than large page, "
+                           "can't use large page\n");
+                return -EINVAL;
+        }
+
+        /* allocate pointer array and alloc large pages */
+        ptrs_size = PFN_ALIGN(num_possible_cpus() * sizeof(pcpur_ptrs[0]));
+        pcpur_ptrs = alloc_bootmem(ptrs_size);
+
+        for_each_possible_cpu(cpu) {
+                pcpur_ptrs[cpu] = pcpu_alloc_bootmem(cpu, PMD_SIZE, PMD_SIZE);
+                if (!pcpur_ptrs[cpu])
+                        goto enomem;
+
+                /*
+                 * Only use pcpur_size bytes and give back the rest.
+                 *
+                 * Ingo: The 2MB up-rounding bootmem is needed to make
+                 * sure the partial 2MB page is still fully RAM - it's
+                 * not well-specified to have a PAT-incompatible area
+                 * (unmapped RAM, device memory, etc.) in that hole.
+                 */
+                free_bootmem(__pa(pcpur_ptrs[cpu] + pcpur_size),
+                             PMD_SIZE - pcpur_size);
+
+                memcpy(pcpur_ptrs[cpu], __per_cpu_load, static_size);
+        }
+
+        /* allocate address and map */
+        vm.flags = VM_ALLOC;
+        vm.size = num_possible_cpus() * PMD_SIZE;
+        vm_area_register_early(&vm, PMD_SIZE);
+
+        for_each_possible_cpu(cpu) {
+                pmd_t *pmd;
+
+                pmd = populate_extra_pmd((unsigned long)vm.addr
+                                         + cpu * PMD_SIZE);
+                set_pmd(pmd, pfn_pmd(page_to_pfn(virt_to_page(pcpur_ptrs[cpu])),
+                                     PAGE_KERNEL_LARGE));
+        }
+
+        /* we're ready, commit */
+        pr_info("PERCPU: Remapped at %p with large pages, static data "
+                "%zu bytes\n", vm.addr, static_size);
+
+        ret = pcpu_setup_first_chunk(pcpur_get_page, static_size, PMD_SIZE,
+                                     pcpur_size - static_size, vm.addr, NULL);
+        goto out_free_ar;
+
+enomem:
+        for_each_possible_cpu(cpu)
+                if (pcpur_ptrs[cpu])
+                        free_bootmem(__pa(pcpur_ptrs[cpu]), PMD_SIZE);
+        ret = -ENOMEM;
+out_free_ar:
+        free_bootmem(__pa(pcpur_ptrs), ptrs_size);
+        return ret;
+}
+#else
+static ssize_t __init setup_pcpu_remap(size_t static_size)
+{
+        return -EINVAL;
+}
+#endif
+
+/*
+ * Embedding allocator
+ *
+ * The first chunk is sized to just contain the static area plus
+ * PERCPU_DYNAMIC_RESERVE and allocated as a contiguous area using
+ * bootmem allocator and used as-is without being mapped into vmalloc
+ * area.  This enables the first chunk to piggy back on the linear
+ * physical PMD mapping and doesn't add any additional pressure to
+ * TLB.
+ */
+static void *pcpue_ptr __initdata;
+static size_t pcpue_unit_size __initdata;
+
+static struct page * __init pcpue_get_page(unsigned int cpu, int pageno)
+{
+        return virt_to_page(pcpue_ptr + cpu * pcpue_unit_size
+                            + ((size_t)pageno << PAGE_SHIFT));
+}
+
+static ssize_t __init setup_pcpu_embed(size_t static_size)
+{
+        unsigned int cpu;
+
+        /*
+         * If large page isn't supported, there's no benefit in doing
+         * this.  Also, embedding allocation doesn't play well with
+         * NUMA.
+         */
+        if (!cpu_has_pse || pcpu_need_numa())
+                return -EINVAL;
+
+        /* allocate and copy */
+        pcpue_unit_size = PFN_ALIGN(static_size + PERCPU_DYNAMIC_RESERVE);
+        pcpue_unit_size = max_t(size_t, pcpue_unit_size, PCPU_MIN_UNIT_SIZE);
+        pcpue_ptr = pcpu_alloc_bootmem(0, num_possible_cpus() * pcpue_unit_size,
+                                       PAGE_SIZE);
+        if (!pcpue_ptr)
+                return -ENOMEM;
+
+        for_each_possible_cpu(cpu)
+                memcpy(pcpue_ptr + cpu * pcpue_unit_size, __per_cpu_load,
+                       static_size);
+
+        /* we're ready, commit */
+        pr_info("PERCPU: Embedded %zu pages at %p, static data %zu bytes\n",
+                pcpue_unit_size >> PAGE_SHIFT, pcpue_ptr, static_size);
+
+        return pcpu_setup_first_chunk(pcpue_get_page, static_size,
+                                      pcpue_unit_size,
+                                      pcpue_unit_size - static_size, pcpue_ptr,
+                                      NULL);
+}
+
+/*
+ * 4k page allocator
+ *
+ * This is the basic allocator.  Static percpu area is allocated
+ * page-by-page and most of initialization is done by the generic
+ * setup function.
+ */
+static struct page **pcpu4k_pages __initdata;
+static int pcpu4k_nr_static_pages __initdata;
+
+static struct page * __init pcpu4k_get_page(unsigned int cpu, int pageno)
+{
+        if (pageno < pcpu4k_nr_static_pages)
+                return pcpu4k_pages[cpu * pcpu4k_nr_static_pages + pageno];
+        return NULL;
+}
+
+static void __init pcpu4k_populate_pte(unsigned long addr)
+{
+        populate_extra_pte(addr);
+}
+
+static ssize_t __init setup_pcpu_4k(size_t static_size)
+{
+        size_t pages_size;
+        unsigned int cpu;
+        int i, j;
+        ssize_t ret;
+
+        pcpu4k_nr_static_pages = PFN_UP(static_size);
+
+        /* unaligned allocations can't be freed, round up to page size */
+        pages_size = PFN_ALIGN(pcpu4k_nr_static_pages * num_possible_cpus()
+                               * sizeof(pcpu4k_pages[0]));
+        pcpu4k_pages = alloc_bootmem(pages_size);
+
+        /* allocate and copy */
+        j = 0;
+        for_each_possible_cpu(cpu)
+                for (i = 0; i < pcpu4k_nr_static_pages; i++) {
+                        void *ptr;
+
+                        ptr = pcpu_alloc_bootmem(cpu, PAGE_SIZE, PAGE_SIZE);
+                        if (!ptr)
+                                goto enomem;
+
+                        memcpy(ptr, __per_cpu_load + i * PAGE_SIZE, PAGE_SIZE);
+                        pcpu4k_pages[j++] = virt_to_page(ptr);
+                }
+
+        /* we're ready, commit */
+        pr_info("PERCPU: Allocated %d 4k pages, static data %zu bytes\n",
+                pcpu4k_nr_static_pages, static_size);
+
+        ret = pcpu_setup_first_chunk(pcpu4k_get_page, static_size, 0, 0, NULL,
+                                     pcpu4k_populate_pte);
+        goto out_free_ar;
+
+enomem:
+        while (--j >= 0)
+                free_bootmem(__pa(page_address(pcpu4k_pages[j])), PAGE_SIZE);
+        ret = -ENOMEM;
+out_free_ar:
+        free_bootmem(__pa(pcpu4k_pages), pages_size);
+        return ret;
+}
+
 static inline void setup_percpu_segment(int cpu)
 {
 #ifdef CONFIG_X86_32
@@ -61,38 +377,35 @@ static inline void setup_percpu_segment(int cpu)
  */
 void __init setup_per_cpu_areas(void)
 {
-        ssize_t size;
-        char *ptr;
-        int cpu;
-
-        /* Copy section for each CPU (we discard the original) */
-        size = roundup(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
+        size_t static_size = __per_cpu_end - __per_cpu_start;
+        unsigned int cpu;
+        unsigned long delta;
+        size_t pcpu_unit_size;
+        ssize_t ret;
 
         pr_info("NR_CPUS:%d nr_cpumask_bits:%d nr_cpu_ids:%d nr_node_ids:%d\n",
                 NR_CPUS, nr_cpumask_bits, nr_cpu_ids, nr_node_ids);
 
-        pr_info("PERCPU: Allocating %zd bytes of per cpu data\n", size);
+        /*
+         * Allocate percpu area.  If PSE is supported, try to make use
+         * of large page mappings.  Please read comments on top of
+         * each allocator for details.
+         */
+        ret = setup_pcpu_remap(static_size);
+        if (ret < 0)
+                ret = setup_pcpu_embed(static_size);
+        if (ret < 0)
+                ret = setup_pcpu_4k(static_size);
+        if (ret < 0)
+                panic("cannot allocate static percpu area (%zu bytes, err=%zd)",
+                      static_size, ret);
 
-        for_each_possible_cpu(cpu) {
-#ifndef CONFIG_NEED_MULTIPLE_NODES
-                ptr = alloc_bootmem_pages(size);
-#else
-                int node = early_cpu_to_node(cpu);
-                if (!node_online(node) || !NODE_DATA(node)) {
-                        ptr = alloc_bootmem_pages(size);
-                        pr_info("cpu %d has no node %d or node-local memory\n",
-                                cpu, node);
-                        pr_debug("per cpu data for cpu%d at %016lx\n",
-                                 cpu, __pa(ptr));
-                } else {
-                        ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
-                        pr_debug("per cpu data for cpu%d on node%d at %016lx\n",
-                                cpu, node, __pa(ptr));
-                }
-#endif
+        pcpu_unit_size = ret;
 
-                memcpy(ptr, __per_cpu_load, __per_cpu_end - __per_cpu_start);
-                per_cpu_offset(cpu) = ptr - __per_cpu_start;
+        /* alrighty, percpu areas up and running */
+        delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
+        for_each_possible_cpu(cpu) {
+                per_cpu_offset(cpu) = delta + cpu * pcpu_unit_size;
                 per_cpu(this_cpu_off, cpu) = per_cpu_offset(cpu);
                 per_cpu(cpu_number, cpu) = cpu;
                 setup_percpu_segment(cpu);
@@ -125,8 +438,6 @@ void __init setup_per_cpu_areas(void)
                  */
                 if (cpu == boot_cpu_id)
                         switch_to_new_gdt(cpu);
-
-                DBG("PERCPU: cpu %4d %p\n", cpu, ptr);
         }
 
         /* indicate the early static arrays will soon be gone */