ceph/src/dpdk/lib/librte_eal/linuxapp/eal/eal_xen_memory.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <errno.h>
  35 #include <stdarg.h>
  36 #include <stdlib.h>
  37 #include <stdio.h>
  38 #include <stdint.h>
  39 #include <inttypes.h>
  40 #include <string.h>
  41 #include <stdarg.h>
  42 #include <sys/mman.h>
  43 #include <sys/types.h>
  44 #include <sys/stat.h>
  45 #include <sys/queue.h>
  46 #include <sys/file.h>
  47 #include <unistd.h>
  48 #include <limits.h>
  49 #include <errno.h>
  50 #include <sys/ioctl.h>
  51 #include <sys/time.h>
  52
  53 #include <rte_log.h>
  54 #include <rte_memory.h>
  55 #include <rte_memzone.h>
  56 #include <rte_launch.h>
  57 #include <rte_eal.h>
  58 #include <rte_eal_memconfig.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_lcore.h>
  61 #include <rte_common.h>
  62 #include <rte_string_fns.h>
  63
  64 #include "eal_private.h"
  65 #include "eal_internal_cfg.h"
  66 #include "eal_filesystem.h"
  67 #include <exec-env/rte_dom0_common.h>
  68
  69 #define PAGE_SIZE RTE_PGSIZE_4K
  70 #define DEFAUL_DOM0_NAME "dom0-mem"
  71
  72 static int xen_fd = -1;
  73 static const char sys_dir_path[] = "/sys/kernel/mm/dom0-mm/memsize-mB";
  74
  75 /*
  76  * Try to mmap *size bytes in /dev/zero. If it is successful, return the
  77  * pointer to the mmap'd area and keep *size unmodified. Else, retry
  78  * with a smaller zone: decrease *size by mem_size until it reaches
  79  * 0. In this case, return NULL. Note: this function returns an address
  80  * which is a multiple of mem_size size.
  81  */
  82 static void *
  83 xen_get_virtual_area(size_t *size, size_t mem_size)
  84 {
  85         void *addr;
  86         int fd;
  87         long aligned_addr;
  88
  89         RTE_LOG(DEBUG, EAL, "Ask a virtual area of 0x%zu bytes\n", *size);
  90
  91         fd = open("/dev/zero", O_RDONLY);
  92         if (fd < 0){
  93                 RTE_LOG(ERR, EAL, "Cannot open /dev/zero\n");
  94                 return NULL;
  95         }
  96         do {
  97                 addr = mmap(NULL, (*size) + mem_size, PROT_READ,
  98                         MAP_PRIVATE, fd, 0);
  99                 if (addr == MAP_FAILED)
 100                         *size -= mem_size;
 101         } while (addr == MAP_FAILED && *size > 0);
 102
 103         if (addr == MAP_FAILED) {
 104                 close(fd);
 105                 RTE_LOG(ERR, EAL, "Cannot get a virtual area\n");
 106                 return NULL;
 107         }
 108
 109         munmap(addr, (*size) + mem_size);
 110         close(fd);
 111
 112         /* align addr to a mem_size boundary */
 113         aligned_addr = (uintptr_t)addr;
 114         aligned_addr = RTE_ALIGN_CEIL(aligned_addr, mem_size);
 115         addr = (void *)(aligned_addr);
 116
 117         RTE_LOG(DEBUG, EAL, "Virtual area found at %p (size = 0x%zx)\n",
 118                 addr, *size);
 119
 120         return addr;
 121 }
 122
 123 /**
 124  * Get memory size configuration from /sys/devices/virtual/misc/dom0_mm
 125  * /memsize-mB/memsize file, and the size unit is mB.
 126  */
 127 static int
 128 get_xen_memory_size(void)
 129 {
 130         char path[PATH_MAX];
 131         unsigned long mem_size = 0;
 132         static const char *file_name;
 133
 134         file_name = "memsize";
 135         snprintf(path, sizeof(path), "%s/%s",
 136                         sys_dir_path, file_name);
 137
 138         if (eal_parse_sysfs_value(path, &mem_size) < 0)
 139                 return -1;
 140
 141         if (mem_size == 0)
 142                 rte_exit(EXIT_FAILURE,"XEN-DOM0:the %s/%s was not"
 143                         " configured.\n",sys_dir_path, file_name);
 144         if (mem_size % 2)
 145                 rte_exit(EXIT_FAILURE,"XEN-DOM0:the %s/%s must be"
 146                         " even number.\n",sys_dir_path, file_name);
 147
 148         if (mem_size > DOM0_CONFIG_MEMSIZE)
 149                 rte_exit(EXIT_FAILURE,"XEN-DOM0:the %s/%s should not be larger"
 150                         " than %d mB\n",sys_dir_path, file_name, DOM0_CONFIG_MEMSIZE);
 151
 152         return mem_size;
 153 }
 154
 155 /**
 156  * Based on physical address to caculate MFN in Xen Dom0.
 157  */
 158 phys_addr_t
 159 rte_xen_mem_phy2mch(int32_t memseg_id, const phys_addr_t phy_addr)
 160 {
 161         int mfn_id, i;
 162         uint64_t mfn, mfn_offset;
 163         struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
 164         struct rte_memseg *memseg = mcfg->memseg;
 165
 166         /* find the memory segment owning the physical address */
 167         if (memseg_id == -1) {
 168                 for (i = 0; i < RTE_MAX_MEMSEG; i++) {
 169                         if ((phy_addr >= memseg[i].phys_addr) &&
 170                                         (phy_addr < memseg[i].phys_addr +
 171                                                 memseg[i].len)) {
 172                                 memseg_id = i;
 173                                 break;
 174                         }
 175                 }
 176                 if (memseg_id == -1)
 177                         return RTE_BAD_PHYS_ADDR;
 178         }
 179
 180         mfn_id = (phy_addr - memseg[memseg_id].phys_addr) / RTE_PGSIZE_2M;
 181
 182         /*the MFN is contiguous in 2M */
 183         mfn_offset = (phy_addr - memseg[memseg_id].phys_addr) %
 184                                         RTE_PGSIZE_2M / PAGE_SIZE;
 185         mfn = mfn_offset + memseg[memseg_id].mfn[mfn_id];
 186
 187         /** return mechine address */
 188         return mfn * PAGE_SIZE + phy_addr % PAGE_SIZE;
 189 }
 190
 191 int
 192 rte_xen_dom0_memory_init(void)
 193 {
 194         void *vir_addr, *vma_addr = NULL;
 195         int err, ret = 0;
 196         uint32_t i, requested, mem_size, memseg_idx, num_memseg = 0;
 197         size_t vma_len = 0;
 198         struct memory_info meminfo;
 199         struct memseg_info seginfo[RTE_MAX_MEMSEG];
 200         int flags, page_size = getpagesize();
 201         struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
 202         struct rte_memseg *memseg = mcfg->memseg;
 203         uint64_t total_mem = internal_config.memory;
 204
 205         memset(seginfo, 0, sizeof(seginfo));
 206         memset(&meminfo, 0, sizeof(struct memory_info));
 207
 208         mem_size = get_xen_memory_size();
 209         requested = (unsigned) (total_mem / 0x100000);
 210         if (requested > mem_size)
 211                 /* if we didn't satisfy total memory requirements */
 212                 rte_exit(EXIT_FAILURE,"Not enough memory available! Requested: %uMB,"
 213                                 " available: %uMB\n", requested, mem_size);
 214         else if (total_mem != 0)
 215                 mem_size = requested;
 216
 217         /* Check FD and open once */
 218         if (xen_fd < 0) {
 219                 xen_fd = open(DOM0_MM_DEV, O_RDWR);
 220                 if (xen_fd < 0) {
 221                         RTE_LOG(ERR, EAL, "Can not open %s\n",DOM0_MM_DEV);
 222                         return -1;
 223                 }
 224         }
 225
 226         meminfo.size = mem_size;
 227
 228         /* construct memory mangement name for Dom0 */
 229         snprintf(meminfo.name, DOM0_NAME_MAX, "%s-%s",
 230                 internal_config.hugefile_prefix, DEFAUL_DOM0_NAME);
 231
 232         /* Notify kernel driver to allocate memory */
 233         ret = ioctl(xen_fd, RTE_DOM0_IOCTL_PREPARE_MEMSEG, &meminfo);
 234         if (ret < 0) {
 235                 RTE_LOG(ERR, EAL, "XEN DOM0:failed to get memory\n");
 236                 err = -EIO;
 237                 goto fail;
 238         }
 239
 240         /* Get number of memory segment from driver */
 241         ret = ioctl(xen_fd, RTE_DOM0_IOCTL_GET_NUM_MEMSEG, &num_memseg);
 242         if (ret < 0) {
 243                 RTE_LOG(ERR, EAL, "XEN DOM0:failed to get memseg count.\n");
 244                 err = -EIO;
 245                 goto fail;
 246         }
 247
 248         if(num_memseg > RTE_MAX_MEMSEG){
 249                 RTE_LOG(ERR, EAL, "XEN DOM0: the memseg count %d is greater"
 250                         " than max memseg %d.\n",num_memseg, RTE_MAX_MEMSEG);
 251                 err = -EIO;
 252                 goto fail;
 253         }
 254
 255         /* get all memory segements information */
 256         ret = ioctl(xen_fd, RTE_DOM0_IOCTL_GET_MEMSEG_INFO, seginfo);
 257         if (ret < 0) {
 258                 RTE_LOG(ERR, EAL, "XEN DOM0:failed to get memseg info.\n");
 259                 err = -EIO;
 260                 goto fail;
 261         }
 262
 263         /* map all memory segments to contiguous user space */
 264         for (memseg_idx = 0; memseg_idx < num_memseg; memseg_idx++)
 265         {
 266                 vma_len = seginfo[memseg_idx].size;
 267
 268                 /**
 269                  * get the biggest virtual memory area up to vma_len. If it fails,
 270                  * vma_addr is NULL, so let the kernel provide the address.
 271                  */
 272                 vma_addr = xen_get_virtual_area(&vma_len, RTE_PGSIZE_2M);
 273                 if (vma_addr == NULL) {
 274                         flags = MAP_SHARED;
 275                         vma_len = RTE_PGSIZE_2M;
 276                 } else
 277                         flags = MAP_SHARED | MAP_FIXED;
 278
 279                 seginfo[memseg_idx].size = vma_len;
 280                 vir_addr = mmap(vma_addr, seginfo[memseg_idx].size,
 281                         PROT_READ|PROT_WRITE, flags, xen_fd,
 282                         memseg_idx * page_size);
 283                 if (vir_addr == MAP_FAILED) {
 284                         RTE_LOG(ERR, EAL, "XEN DOM0:Could not mmap %s\n",
 285                                 DOM0_MM_DEV);
 286                         err = -EIO;
 287                         goto fail;
 288                 }
 289
 290                 memseg[memseg_idx].addr = vir_addr;
 291                 memseg[memseg_idx].phys_addr = page_size *
 292                         seginfo[memseg_idx].pfn ;
 293                 memseg[memseg_idx].len = seginfo[memseg_idx].size;
 294                 for ( i = 0; i < seginfo[memseg_idx].size / RTE_PGSIZE_2M; i++)
 295                         memseg[memseg_idx].mfn[i] = seginfo[memseg_idx].mfn[i];
 296
 297                 /* MFNs are continuous in 2M, so assume that page size is 2M */
 298                 memseg[memseg_idx].hugepage_sz = RTE_PGSIZE_2M;
 299
 300                 memseg[memseg_idx].nchannel = mcfg->nchannel;
 301                 memseg[memseg_idx].nrank = mcfg->nrank;
 302
 303                 /* NUMA is not suppoted in Xen Dom0, so only set socket 0*/
 304                 memseg[memseg_idx].socket_id = 0;
 305         }
 306
 307         return 0;
 308 fail:
 309         if (xen_fd > 0) {
 310                 close(xen_fd);
 311                 xen_fd = -1;
 312         }
 313         return err;
 314 }
 315
 316 /*
 317  * This creates the memory mappings in the secondary process to match that of
 318  * the server process. It goes through each memory segment in the DPDK runtime
 319  * configuration, mapping them in order to form a contiguous block in the
 320  * virtual memory space
 321  */
 322 int
 323 rte_xen_dom0_memory_attach(void)
 324 {
 325         const struct rte_mem_config *mcfg;
 326         unsigned s = 0; /* s used to track the segment number */
 327         int xen_fd = -1;
 328         int ret = -1;
 329         void *vir_addr;
 330         char name[DOM0_NAME_MAX] = {0};
 331         int page_size = getpagesize();
 332
 333         mcfg = rte_eal_get_configuration()->mem_config;
 334
 335         /* Check FD and open once */
 336         if (xen_fd < 0) {
 337                 xen_fd = open(DOM0_MM_DEV, O_RDWR);
 338                 if (xen_fd < 0) {
 339                         RTE_LOG(ERR, EAL, "Can not open %s\n",DOM0_MM_DEV);
 340                         goto error;
 341                 }
 342         }
 343
 344         /* construct memory mangement name for Dom0 */
 345         snprintf(name, DOM0_NAME_MAX, "%s-%s",
 346                 internal_config.hugefile_prefix, DEFAUL_DOM0_NAME);
 347         /* attach to memory segments of primary process */
 348         ret = ioctl(xen_fd, RTE_DOM0_IOCTL_ATTACH_TO_MEMSEG, name);
 349         if (ret) {
 350                 RTE_LOG(ERR, EAL,"attach memory segments fail.\n");
 351                 goto error;
 352         }
 353
 354         /* map all segments into memory to make sure we get the addrs */
 355         for (s = 0; s < RTE_MAX_MEMSEG; ++s) {
 356
 357                 /*
 358                  * the first memory segment with len==0 is the one that
 359                  * follows the last valid segment.
 360                  */
 361                 if (mcfg->memseg[s].len == 0)
 362                         break;
 363
 364                 vir_addr = mmap(mcfg->memseg[s].addr, mcfg->memseg[s].len,
 365                                 PROT_READ|PROT_WRITE, MAP_SHARED|MAP_FIXED, xen_fd,
 366                                 s * page_size);
 367                 if (vir_addr == MAP_FAILED) {
 368                         RTE_LOG(ERR, EAL, "Could not mmap %llu bytes "
 369                                 "in %s to requested address [%p]\n",
 370                                 (unsigned long long)mcfg->memseg[s].len, DOM0_MM_DEV,
 371                                 mcfg->memseg[s].addr);
 372                         goto error;
 373                 }
 374         }
 375         return 0;
 376
 377 error:
 378         if (xen_fd >= 0) {
 379                 close(xen_fd);
 380                 xen_fd = -1;
 381         }
 382         return -1;
 383 }