drivers/xen/unpopulated-alloc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/errno.h>
   3 #include <linux/gfp.h>
   4 #include <linux/kernel.h>
   5 #include <linux/mm.h>
   6 #include <linux/memremap.h>
   7 #include <linux/slab.h>
   8
   9 #include <asm/page.h>
  10
  11 #include <xen/page.h>
  12 #include <xen/xen.h>
  13
  14 static DEFINE_MUTEX(list_lock);
  15 static struct page *page_list;
  16 static unsigned int list_count;
  17
  18 static int fill_list(unsigned int nr_pages)
  19 {
  20         struct dev_pagemap *pgmap;
  21         struct resource *res;
  22         void *vaddr;
  23         unsigned int i, alloc_pages = round_up(nr_pages, PAGES_PER_SECTION);
  24         int ret = -ENOMEM;
  25
  26         res = kzalloc(sizeof(*res), GFP_KERNEL);
  27         if (!res)
  28                 return -ENOMEM;
  29
  30         res->name = "Xen scratch";
  31         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  32
  33         ret = allocate_resource(&iomem_resource, res,
  34                                 alloc_pages * PAGE_SIZE, 0, -1,
  35                                 PAGES_PER_SECTION * PAGE_SIZE, NULL, NULL);
  36         if (ret < 0) {
  37                 pr_err("Cannot allocate new IOMEM resource\n");
  38                 goto err_resource;
  39         }
  40
  41         pgmap = kzalloc(sizeof(*pgmap), GFP_KERNEL);
  42         if (!pgmap)
  43                 goto err_pgmap;
  44
  45         pgmap->type = MEMORY_DEVICE_GENERIC;
  46         pgmap->range = (struct range) {
  47                 .start = res->start,
  48                 .end = res->end,
  49         };
  50         pgmap->nr_range = 1;
  51         pgmap->owner = res;
  52
  53 #ifdef CONFIG_XEN_HAVE_PVMMU
  54         /*
  55          * memremap will build page tables for the new memory so
  56          * the p2m must contain invalid entries so the correct
  57          * non-present PTEs will be written.
  58          *
  59          * If a failure occurs, the original (identity) p2m entries
  60          * are not restored since this region is now known not to
  61          * conflict with any devices.
  62          */
  63         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
  64                 xen_pfn_t pfn = PFN_DOWN(res->start);
  65
  66                 for (i = 0; i < alloc_pages; i++) {
  67                         if (!set_phys_to_machine(pfn + i, INVALID_P2M_ENTRY)) {
  68                                 pr_warn("set_phys_to_machine() failed, no memory added\n");
  69                                 ret = -ENOMEM;
  70                                 goto err_memremap;
  71                         }
  72                 }
  73         }
  74 #endif
  75
  76         vaddr = memremap_pages(pgmap, NUMA_NO_NODE);
  77         if (IS_ERR(vaddr)) {
  78                 pr_err("Cannot remap memory range\n");
  79                 ret = PTR_ERR(vaddr);
  80                 goto err_memremap;
  81         }
  82
  83         for (i = 0; i < alloc_pages; i++) {
  84                 struct page *pg = virt_to_page(vaddr + PAGE_SIZE * i);
  85
  86                 BUG_ON(!virt_addr_valid(vaddr + PAGE_SIZE * i));
  87                 pg->zone_device_data = page_list;
  88                 page_list = pg;
  89                 list_count++;
  90         }
  91
  92         return 0;
  93
  94 err_memremap:
  95         kfree(pgmap);
  96 err_pgmap:
  97         release_resource(res);
  98 err_resource:
  99         kfree(res);
 100         return ret;
 101 }
 102
 103 /**
 104  * xen_alloc_unpopulated_pages - alloc unpopulated pages
 105  * @nr_pages: Number of pages
 106  * @pages: pages returned
 107  * @return 0 on success, error otherwise
 108  */
 109 int xen_alloc_unpopulated_pages(unsigned int nr_pages, struct page **pages)
 110 {
 111         unsigned int i;
 112         int ret = 0;
 113
 114         mutex_lock(&list_lock);
 115         if (list_count < nr_pages) {
 116                 ret = fill_list(nr_pages - list_count);
 117                 if (ret)
 118                         goto out;
 119         }
 120
 121         for (i = 0; i < nr_pages; i++) {
 122                 struct page *pg = page_list;
 123
 124                 BUG_ON(!pg);
 125                 page_list = pg->zone_device_data;
 126                 list_count--;
 127                 pages[i] = pg;
 128
 129 #ifdef CONFIG_XEN_HAVE_PVMMU
 130                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 131                         ret = xen_alloc_p2m_entry(page_to_pfn(pg));
 132                         if (ret < 0) {
 133                                 unsigned int j;
 134
 135                                 for (j = 0; j <= i; j++) {
 136                                         pages[j]->zone_device_data = page_list;
 137                                         page_list = pages[j];
 138                                         list_count++;
 139                                 }
 140                                 goto out;
 141                         }
 142                 }
 143 #endif
 144         }
 145
 146 out:
 147         mutex_unlock(&list_lock);
 148         return ret;
 149 }
 150 EXPORT_SYMBOL(xen_alloc_unpopulated_pages);
 151
 152 /**
 153  * xen_free_unpopulated_pages - return unpopulated pages
 154  * @nr_pages: Number of pages
 155  * @pages: pages to return
 156  */
 157 void xen_free_unpopulated_pages(unsigned int nr_pages, struct page **pages)
 158 {
 159         unsigned int i;
 160
 161         mutex_lock(&list_lock);
 162         for (i = 0; i < nr_pages; i++) {
 163                 pages[i]->zone_device_data = page_list;
 164                 page_list = pages[i];
 165                 list_count++;
 166         }
 167         mutex_unlock(&list_lock);
 168 }
 169 EXPORT_SYMBOL(xen_free_unpopulated_pages);
 170
 171 #ifdef CONFIG_XEN_PV
 172 static int __init init(void)
 173 {
 174         unsigned int i;
 175
 176         if (!xen_domain())
 177                 return -ENODEV;
 178
 179         if (!xen_pv_domain())
 180                 return 0;
 181
 182         /*
 183          * Initialize with pages from the extra memory regions (see
 184          * arch/x86/xen/setup.c).
 185          */
 186         for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
 187                 unsigned int j;
 188
 189                 for (j = 0; j < xen_extra_mem[i].n_pfns; j++) {
 190                         struct page *pg =
 191                                 pfn_to_page(xen_extra_mem[i].start_pfn + j);
 192
 193                         pg->zone_device_data = page_list;
 194                         page_list = pg;
 195                         list_count++;
 196                 }
 197         }
 198
 199         return 0;
 200 }
 201 subsys_initcall(init);
 202 #endif