drivers/dax/kmem.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 2016-2019 Intel Corporation. All rights reserved. */
   3 #include <linux/memremap.h>
   4 #include <linux/pagemap.h>
   5 #include <linux/memory.h>
   6 #include <linux/module.h>
   7 #include <linux/device.h>
   8 #include <linux/pfn_t.h>
   9 #include <linux/slab.h>
  10 #include <linux/dax.h>
  11 #include <linux/fs.h>
  12 #include <linux/mm.h>
  13 #include <linux/mman.h>
  14 #include "dax-private.h"
  15 #include "bus.h"
  16
  17 /* Memory resource name used for add_memory_driver_managed(). */
  18 static const char *kmem_name;
  19 /* Set if any memory will remain added when the driver will be unloaded. */
  20 static bool any_hotremove_failed;
  21
  22 static int dax_kmem_range(struct dev_dax *dev_dax, int i, struct range *r)
  23 {
  24         struct dev_dax_range *dax_range = &dev_dax->ranges[i];
  25         struct range *range = &dax_range->range;
  26
  27         /* memory-block align the hotplug range */
  28         r->start = ALIGN(range->start, memory_block_size_bytes());
  29         r->end = ALIGN_DOWN(range->end + 1, memory_block_size_bytes()) - 1;
  30         if (r->start >= r->end) {
  31                 r->start = range->start;
  32                 r->end = range->end;
  33                 return -ENOSPC;
  34         }
  35         return 0;
  36 }
  37
  38 struct dax_kmem_data {
  39         const char *res_name;
  40         int mgid;
  41         struct resource *res[];
  42 };
  43
  44 static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
  45 {
  46         struct device *dev = &dev_dax->dev;
  47         unsigned long total_len = 0;
  48         struct dax_kmem_data *data;
  49         int i, rc, mapped = 0;
  50         int numa_node;
  51
  52         /*
  53          * Ensure good NUMA information for the persistent memory.
  54          * Without this check, there is a risk that slow memory
  55          * could be mixed in a node with faster memory, causing
  56          * unavoidable performance issues.
  57          */
  58         numa_node = dev_dax->target_node;
  59         if (numa_node < 0) {
  60                 dev_warn(dev, "rejecting DAX region with invalid node: %d\n",
  61                                 numa_node);
  62                 return -EINVAL;
  63         }
  64
  65         for (i = 0; i < dev_dax->nr_range; i++) {
  66                 struct range range;
  67
  68                 rc = dax_kmem_range(dev_dax, i, &range);
  69                 if (rc) {
  70                         dev_info(dev, "mapping%d: %#llx-%#llx too small after alignment\n",
  71                                         i, range.start, range.end);
  72                         continue;
  73                 }
  74                 total_len += range_len(&range);
  75         }
  76
  77         if (!total_len) {
  78                 dev_warn(dev, "rejecting DAX region without any memory after alignment\n");
  79                 return -EINVAL;
  80         }
  81
  82         data = kzalloc(struct_size(data, res, dev_dax->nr_range), GFP_KERNEL);
  83         if (!data)
  84                 return -ENOMEM;
  85
  86         rc = -ENOMEM;
  87         data->res_name = kstrdup(dev_name(dev), GFP_KERNEL);
  88         if (!data->res_name)
  89                 goto err_res_name;
  90
  91         rc = memory_group_register_static(numa_node, total_len);
  92         if (rc < 0)
  93                 goto err_reg_mgid;
  94         data->mgid = rc;
  95
  96         for (i = 0; i < dev_dax->nr_range; i++) {
  97                 struct resource *res;
  98                 struct range range;
  99
 100                 rc = dax_kmem_range(dev_dax, i, &range);
 101                 if (rc)
 102                         continue;
 103
 104                 /* Region is permanently reserved if hotremove fails. */
 105                 res = request_mem_region(range.start, range_len(&range), data->res_name);
 106                 if (!res) {
 107                         dev_warn(dev, "mapping%d: %#llx-%#llx could not reserve region\n",
 108                                         i, range.start, range.end);
 109                         /*
 110                          * Once some memory has been onlined we can't
 111                          * assume that it can be un-onlined safely.
 112                          */
 113                         if (mapped)
 114                                 continue;
 115                         rc = -EBUSY;
 116                         goto err_request_mem;
 117                 }
 118                 data->res[i] = res;
 119
 120                 /*
 121                  * Set flags appropriate for System RAM.  Leave ..._BUSY clear
 122                  * so that add_memory() can add a child resource.  Do not
 123                  * inherit flags from the parent since it may set new flags
 124                  * unknown to us that will break add_memory() below.
 125                  */
 126                 res->flags = IORESOURCE_SYSTEM_RAM;
 127
 128                 /*
 129                  * Ensure that future kexec'd kernels will not treat
 130                  * this as RAM automatically.
 131                  */
 132                 rc = add_memory_driver_managed(data->mgid, range.start,
 133                                 range_len(&range), kmem_name, MHP_NID_IS_MGID);
 134
 135                 if (rc) {
 136                         dev_warn(dev, "mapping%d: %#llx-%#llx memory add failed\n",
 137                                         i, range.start, range.end);
 138                         release_resource(res);
 139                         kfree(res);
 140                         data->res[i] = NULL;
 141                         if (mapped)
 142                                 continue;
 143                         goto err_request_mem;
 144                 }
 145                 mapped++;
 146         }
 147
 148         dev_set_drvdata(dev, data);
 149
 150         return 0;
 151
 152 err_request_mem:
 153         memory_group_unregister(data->mgid);
 154 err_reg_mgid:
 155         kfree(data->res_name);
 156 err_res_name:
 157         kfree(data);
 158         return rc;
 159 }
 160
 161 #ifdef CONFIG_MEMORY_HOTREMOVE
 162 static void dev_dax_kmem_remove(struct dev_dax *dev_dax)
 163 {
 164         int i, success = 0;
 165         struct device *dev = &dev_dax->dev;
 166         struct dax_kmem_data *data = dev_get_drvdata(dev);
 167
 168         /*
 169          * We have one shot for removing memory, if some memory blocks were not
 170          * offline prior to calling this function remove_memory() will fail, and
 171          * there is no way to hotremove this memory until reboot because device
 172          * unbind will succeed even if we return failure.
 173          */
 174         for (i = 0; i < dev_dax->nr_range; i++) {
 175                 struct range range;
 176                 int rc;
 177
 178                 rc = dax_kmem_range(dev_dax, i, &range);
 179                 if (rc)
 180                         continue;
 181
 182                 rc = remove_memory(range.start, range_len(&range));
 183                 if (rc == 0) {
 184                         release_resource(data->res[i]);
 185                         kfree(data->res[i]);
 186                         data->res[i] = NULL;
 187                         success++;
 188                         continue;
 189                 }
 190                 any_hotremove_failed = true;
 191                 dev_err(dev,
 192                         "mapping%d: %#llx-%#llx cannot be hotremoved until the next reboot\n",
 193                                 i, range.start, range.end);
 194         }
 195
 196         if (success >= dev_dax->nr_range) {
 197                 memory_group_unregister(data->mgid);
 198                 kfree(data->res_name);
 199                 kfree(data);
 200                 dev_set_drvdata(dev, NULL);
 201         }
 202 }
 203 #else
 204 static void dev_dax_kmem_remove(struct dev_dax *dev_dax)
 205 {
 206         /*
 207          * Without hotremove purposely leak the request_mem_region() for the
 208          * device-dax range and return '0' to ->remove() attempts. The removal
 209          * of the device from the driver always succeeds, but the region is
 210          * permanently pinned as reserved by the unreleased
 211          * request_mem_region().
 212          */
 213         any_hotremove_failed = true;
 214 }
 215 #endif /* CONFIG_MEMORY_HOTREMOVE */
 216
 217 static struct dax_device_driver device_dax_kmem_driver = {
 218         .probe = dev_dax_kmem_probe,
 219         .remove = dev_dax_kmem_remove,
 220 };
 221
 222 static int __init dax_kmem_init(void)
 223 {
 224         int rc;
 225
 226         /* Resource name is permanently allocated if any hotremove fails. */
 227         kmem_name = kstrdup_const("System RAM (kmem)", GFP_KERNEL);
 228         if (!kmem_name)
 229                 return -ENOMEM;
 230
 231         rc = dax_driver_register(&device_dax_kmem_driver);
 232         if (rc)
 233                 kfree_const(kmem_name);
 234         return rc;
 235 }
 236
 237 static void __exit dax_kmem_exit(void)
 238 {
 239         dax_driver_unregister(&device_dax_kmem_driver);
 240         if (!any_hotremove_failed)
 241                 kfree_const(kmem_name);
 242 }
 243
 244 MODULE_AUTHOR("Intel Corporation");
 245 MODULE_LICENSE("GPL v2");
 246 module_init(dax_kmem_init);
 247 module_exit(dax_kmem_exit);
 248 MODULE_ALIAS_DAX_DEVICE(0);