4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
35 * Copyright(c) 2013 6WIND.
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
41 * * Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * * Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in
45 * the documentation and/or other materials provided with the
47 * * Neither the name of 6WIND S.A. nor the names of its
48 * contributors may be used to endorse or promote products derived
49 * from this software without specific prior written permission.
51 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
52 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
53 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
54 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
55 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
56 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
57 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
58 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
59 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
60 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
61 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
64 #define _FILE_OFFSET_BITS 64
74 #include <sys/types.h>
76 #include <sys/queue.h>
81 #include <sys/ioctl.h>
87 #include <rte_memory.h>
88 #include <rte_memzone.h>
89 #include <rte_launch.h>
91 #include <rte_eal_memconfig.h>
92 #include <rte_per_lcore.h>
93 #include <rte_lcore.h>
94 #include <rte_common.h>
95 #include <rte_string_fns.h>
97 #include "eal_private.h"
98 #include "eal_internal_cfg.h"
99 #include "eal_filesystem.h"
100 #include "eal_hugepages.h"
102 #ifdef RTE_LIBRTE_XEN_DOM0
103 int rte_xen_dom0_supported(void)
105 return internal_config.xen_dom0_support;
111 * Huge page mapping under linux
113 * To reserve a big contiguous amount of memory, we use the hugepage
114 * feature of linux. For that, we need to have hugetlbfs mounted. This
115 * code will create many files in this directory (one per page) and
116 * map them in virtual memory. For each page, we will retrieve its
117 * physical address and remap it in order to have a virtual contiguous
118 * zone as well as a physical contiguous zone.
121 static uint64_t baseaddr_offset;
123 static unsigned proc_pagemap_readable;
125 #define RANDOMIZE_VA_SPACE_FILE "/proc/sys/kernel/randomize_va_space"
128 test_proc_pagemap_readable(void)
130 int fd = open("/proc/self/pagemap", O_RDONLY);
134 "Cannot open /proc/self/pagemap: %s. "
135 "virt2phys address translation will not work\n",
142 proc_pagemap_readable = 1;
145 /* Lock page in physical memory and prevent from swapping. */
147 rte_mem_lock_page(const void *virt)
149 unsigned long virtual = (unsigned long)virt;
150 int page_size = getpagesize();
151 unsigned long aligned = (virtual & ~ (page_size - 1));
152 return mlock((void*)aligned, page_size);
156 * Get physical address of any mapped virtual address in the current process.
159 rte_mem_virt2phy(const void *virtaddr)
162 uint64_t page, physaddr;
163 unsigned long virt_pfn;
167 /* when using dom0, /proc/self/pagemap always returns 0, check in
168 * dpdk memory by browsing the memsegs */
169 if (rte_xen_dom0_supported()) {
170 struct rte_mem_config *mcfg;
171 struct rte_memseg *memseg;
174 mcfg = rte_eal_get_configuration()->mem_config;
175 for (i = 0; i < RTE_MAX_MEMSEG; i++) {
176 memseg = &mcfg->memseg[i];
177 if (memseg->addr == NULL)
179 if (virtaddr > memseg->addr &&
180 virtaddr < RTE_PTR_ADD(memseg->addr,
182 return memseg->phys_addr +
183 RTE_PTR_DIFF(virtaddr, memseg->addr);
187 return RTE_BAD_PHYS_ADDR;
190 /* Cannot parse /proc/self/pagemap, no need to log errors everywhere */
191 if (!proc_pagemap_readable)
192 return RTE_BAD_PHYS_ADDR;
194 /* standard page size */
195 page_size = getpagesize();
197 fd = open("/proc/self/pagemap", O_RDONLY);
199 RTE_LOG(ERR, EAL, "%s(): cannot open /proc/self/pagemap: %s\n",
200 __func__, strerror(errno));
201 return RTE_BAD_PHYS_ADDR;
204 virt_pfn = (unsigned long)virtaddr / page_size;
205 offset = sizeof(uint64_t) * virt_pfn;
206 if (lseek(fd, offset, SEEK_SET) == (off_t) -1) {
207 RTE_LOG(ERR, EAL, "%s(): seek error in /proc/self/pagemap: %s\n",
208 __func__, strerror(errno));
210 return RTE_BAD_PHYS_ADDR;
212 if (read(fd, &page, sizeof(uint64_t)) < 0) {
213 RTE_LOG(ERR, EAL, "%s(): cannot read /proc/self/pagemap: %s\n",
214 __func__, strerror(errno));
216 return RTE_BAD_PHYS_ADDR;
220 * the pfn (page frame number) are bits 0-54 (see
221 * pagemap.txt in linux Documentation)
223 physaddr = ((page & 0x7fffffffffffffULL) * page_size)
224 + ((unsigned long)virtaddr % page_size);
230 * For each hugepage in hugepg_tbl, fill the physaddr value. We find
231 * it by browsing the /proc/self/pagemap special file.
234 find_physaddrs(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi)
239 for (i = 0; i < hpi->num_pages[0]; i++) {
240 addr = rte_mem_virt2phy(hugepg_tbl[i].orig_va);
241 if (addr == RTE_BAD_PHYS_ADDR)
243 hugepg_tbl[i].physaddr = addr;
249 * Check whether address-space layout randomization is enabled in
250 * the kernel. This is important for multi-process as it can prevent
251 * two processes mapping data to the same virtual address
253 * 0 - address space randomization disabled
254 * 1/2 - address space randomization enabled
255 * negative error code on error
261 int retval, fd = open(RANDOMIZE_VA_SPACE_FILE, O_RDONLY);
264 retval = read(fd, &c, 1);
274 default: return -EINVAL;
279 * Try to mmap *size bytes in /dev/zero. If it is successful, return the
280 * pointer to the mmap'd area and keep *size unmodified. Else, retry
281 * with a smaller zone: decrease *size by hugepage_sz until it reaches
282 * 0. In this case, return NULL. Note: this function returns an address
283 * which is a multiple of hugepage size.
286 get_virtual_area(size_t *size, size_t hugepage_sz)
292 if (internal_config.base_virtaddr != 0) {
293 addr = (void*) (uintptr_t) (internal_config.base_virtaddr +
298 RTE_LOG(DEBUG, EAL, "Ask a virtual area of 0x%zx bytes\n", *size);
300 fd = open("/dev/zero", O_RDONLY);
302 RTE_LOG(ERR, EAL, "Cannot open /dev/zero\n");
307 (*size) + hugepage_sz, PROT_READ, MAP_PRIVATE, fd, 0);
308 if (addr == MAP_FAILED)
309 *size -= hugepage_sz;
310 } while (addr == MAP_FAILED && *size > 0);
312 if (addr == MAP_FAILED) {
314 RTE_LOG(ERR, EAL, "Cannot get a virtual area: %s\n",
319 munmap(addr, (*size) + hugepage_sz);
322 /* align addr to a huge page size boundary */
323 aligned_addr = (long)addr;
324 aligned_addr += (hugepage_sz - 1);
325 aligned_addr &= (~(hugepage_sz - 1));
326 addr = (void *)(aligned_addr);
328 RTE_LOG(DEBUG, EAL, "Virtual area found at %p (size = 0x%zx)\n",
331 /* increment offset */
332 baseaddr_offset += *size;
337 static sigjmp_buf huge_jmpenv;
339 static void huge_sigbus_handler(int signo __rte_unused)
341 siglongjmp(huge_jmpenv, 1);
344 /* Put setjmp into a wrap method to avoid compiling error. Any non-volatile,
345 * non-static local variable in the stack frame calling sigsetjmp might be
346 * clobbered by a call to longjmp.
348 static int huge_wrap_sigsetjmp(void)
350 return sigsetjmp(huge_jmpenv, 1);
354 * Mmap all hugepages of hugepage table: it first open a file in
355 * hugetlbfs, then mmap() hugepage_sz data in it. If orig is set, the
356 * virtual address is stored in hugepg_tbl[i].orig_va, else it is stored
357 * in hugepg_tbl[i].final_va. The second mapping (when orig is 0) tries to
358 * map continguous physical blocks in contiguous virtual blocks.
361 map_all_hugepages(struct hugepage_file *hugepg_tbl,
362 struct hugepage_info *hpi, int orig)
367 void *vma_addr = NULL;
370 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
371 RTE_SET_USED(vma_len);
374 for (i = 0; i < hpi->num_pages[0]; i++) {
375 uint64_t hugepage_sz = hpi->hugepage_sz;
378 hugepg_tbl[i].file_id = i;
379 hugepg_tbl[i].size = hugepage_sz;
380 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
381 eal_get_hugefile_temp_path(hugepg_tbl[i].filepath,
382 sizeof(hugepg_tbl[i].filepath), hpi->hugedir,
383 hugepg_tbl[i].file_id);
385 eal_get_hugefile_path(hugepg_tbl[i].filepath,
386 sizeof(hugepg_tbl[i].filepath), hpi->hugedir,
387 hugepg_tbl[i].file_id);
389 hugepg_tbl[i].filepath[sizeof(hugepg_tbl[i].filepath) - 1] = '\0';
392 /* for 32-bit systems, don't remap 1G and 16G pages, just reuse
393 * original map address as final map address.
395 else if ((hugepage_sz == RTE_PGSIZE_1G)
396 || (hugepage_sz == RTE_PGSIZE_16G)) {
397 hugepg_tbl[i].final_va = hugepg_tbl[i].orig_va;
398 hugepg_tbl[i].orig_va = NULL;
403 #ifndef RTE_EAL_SINGLE_FILE_SEGMENTS
404 else if (vma_len == 0) {
405 unsigned j, num_pages;
407 /* reserve a virtual area for next contiguous
408 * physical block: count the number of
409 * contiguous physical pages. */
410 for (j = i+1; j < hpi->num_pages[0] ; j++) {
411 #ifdef RTE_ARCH_PPC_64
412 /* The physical addresses are sorted in
413 * descending order on PPC64 */
414 if (hugepg_tbl[j].physaddr !=
415 hugepg_tbl[j-1].physaddr - hugepage_sz)
418 if (hugepg_tbl[j].physaddr !=
419 hugepg_tbl[j-1].physaddr + hugepage_sz)
424 vma_len = num_pages * hugepage_sz;
426 /* get the biggest virtual memory area up to
427 * vma_len. If it fails, vma_addr is NULL, so
428 * let the kernel provide the address. */
429 vma_addr = get_virtual_area(&vma_len, hpi->hugepage_sz);
430 if (vma_addr == NULL)
431 vma_len = hugepage_sz;
435 /* try to create hugepage file */
436 fd = open(hugepg_tbl[i].filepath, O_CREAT | O_RDWR, 0755);
438 RTE_LOG(DEBUG, EAL, "%s(): open failed: %s\n", __func__,
443 /* map the segment, and populate page tables,
444 * the kernel fills this segment with zeros */
445 virtaddr = mmap(vma_addr, hugepage_sz, PROT_READ | PROT_WRITE,
446 MAP_SHARED | MAP_POPULATE, fd, 0);
447 if (virtaddr == MAP_FAILED) {
448 RTE_LOG(DEBUG, EAL, "%s(): mmap failed: %s\n", __func__,
455 hugepg_tbl[i].orig_va = virtaddr;
458 hugepg_tbl[i].final_va = virtaddr;
462 /* In linux, hugetlb limitations, like cgroup, are
463 * enforced at fault time instead of mmap(), even
464 * with the option of MAP_POPULATE. Kernel will send
465 * a SIGBUS signal. To avoid to be killed, save stack
466 * environment here, if SIGBUS happens, we can jump
469 if (huge_wrap_sigsetjmp()) {
470 RTE_LOG(DEBUG, EAL, "SIGBUS: Cannot mmap more "
471 "hugepages of size %u MB\n",
472 (unsigned)(hugepage_sz / 0x100000));
473 munmap(virtaddr, hugepage_sz);
475 unlink(hugepg_tbl[i].filepath);
478 *(int *)virtaddr = 0;
482 /* set shared flock on the file. */
483 if (flock(fd, LOCK_SH | LOCK_NB) == -1) {
484 RTE_LOG(DEBUG, EAL, "%s(): Locking file failed:%s \n",
485 __func__, strerror(errno));
492 vma_addr = (char *)vma_addr + hugepage_sz;
493 vma_len -= hugepage_sz;
499 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
502 * Remaps all hugepages into single file segments
505 remap_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi)
508 unsigned i = 0, j, num_pages, page_idx = 0;
509 void *vma_addr = NULL, *old_addr = NULL, *page_addr = NULL;
511 size_t hugepage_sz = hpi->hugepage_sz;
512 size_t total_size, offset;
513 char filepath[MAX_HUGEPAGE_PATH];
514 phys_addr_t physaddr;
517 while (i < hpi->num_pages[0]) {
520 /* for 32-bit systems, don't remap 1G pages and 16G pages,
521 * just reuse original map address as final map address.
523 if ((hugepage_sz == RTE_PGSIZE_1G)
524 || (hugepage_sz == RTE_PGSIZE_16G)) {
525 hugepg_tbl[i].final_va = hugepg_tbl[i].orig_va;
526 hugepg_tbl[i].orig_va = NULL;
532 /* reserve a virtual area for next contiguous
533 * physical block: count the number of
534 * contiguous physical pages. */
535 for (j = i+1; j < hpi->num_pages[0] ; j++) {
536 #ifdef RTE_ARCH_PPC_64
537 /* The physical addresses are sorted in descending
539 if (hugepg_tbl[j].physaddr !=
540 hugepg_tbl[j-1].physaddr - hugepage_sz)
543 if (hugepg_tbl[j].physaddr !=
544 hugepg_tbl[j-1].physaddr + hugepage_sz)
549 vma_len = num_pages * hugepage_sz;
551 socket = hugepg_tbl[i].socket_id;
553 /* get the biggest virtual memory area up to
554 * vma_len. If it fails, vma_addr is NULL, so
555 * let the kernel provide the address. */
556 vma_addr = get_virtual_area(&vma_len, hpi->hugepage_sz);
558 /* If we can't find a big enough virtual area, work out how many pages
559 * we are going to get */
560 if (vma_addr == NULL)
562 else if (vma_len != num_pages * hugepage_sz) {
563 num_pages = vma_len / hugepage_sz;
568 hugepg_tbl[page_idx].file_id = page_idx;
569 eal_get_hugefile_path(filepath,
572 hugepg_tbl[page_idx].file_id);
574 /* try to create hugepage file */
575 fd = open(filepath, O_CREAT | O_RDWR, 0755);
577 RTE_LOG(ERR, EAL, "%s(): open failed: %s\n", __func__, strerror(errno));
584 /* unmap current segment */
586 munmap(vma_addr, total_size);
588 /* unmap original page */
589 munmap(hugepg_tbl[i].orig_va, hugepage_sz);
590 unlink(hugepg_tbl[i].filepath);
592 total_size += hugepage_sz;
596 /* map new, bigger segment, and populate page tables,
597 * the kernel fills this segment with zeros */
598 vma_addr = mmap(vma_addr, total_size,
599 PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, 0);
601 if (vma_addr == MAP_FAILED || vma_addr != old_addr) {
602 RTE_LOG(ERR, EAL, "%s(): mmap failed: %s\n", __func__, strerror(errno));
608 /* set shared flock on the file. */
609 if (flock(fd, LOCK_SH | LOCK_NB) == -1) {
610 RTE_LOG(ERR, EAL, "%s(): Locking file failed:%s \n",
611 __func__, strerror(errno));
616 snprintf(hugepg_tbl[page_idx].filepath, MAX_HUGEPAGE_PATH, "%s",
619 physaddr = rte_mem_virt2phy(vma_addr);
621 if (physaddr == RTE_BAD_PHYS_ADDR)
624 hugepg_tbl[page_idx].final_va = vma_addr;
626 hugepg_tbl[page_idx].physaddr = physaddr;
628 hugepg_tbl[page_idx].repeated = num_pages;
630 hugepg_tbl[page_idx].socket_id = socket;
634 /* verify the memory segment - that is, check that every VA corresponds
635 * to the physical address we expect to see
637 for (offset = 0; offset < vma_len; offset += hugepage_sz) {
638 uint64_t expected_physaddr;
640 expected_physaddr = hugepg_tbl[page_idx].physaddr + offset;
641 page_addr = RTE_PTR_ADD(vma_addr, offset);
642 physaddr = rte_mem_virt2phy(page_addr);
644 if (physaddr != expected_physaddr) {
645 RTE_LOG(ERR, EAL, "Segment sanity check failed: wrong physaddr "
646 "at %p (offset 0x%" PRIx64 ": 0x%" PRIx64
647 " (expected 0x%" PRIx64 ")\n",
648 page_addr, offset, physaddr, expected_physaddr);
656 /* zero out the rest */
657 memset(&hugepg_tbl[page_idx], 0, (hpi->num_pages[0] - page_idx) * sizeof(struct hugepage_file));
660 #else/* RTE_EAL_SINGLE_FILE_SEGMENTS=n */
662 /* Unmap all hugepages from original mapping */
664 unmap_all_hugepages_orig(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi)
667 for (i = 0; i < hpi->num_pages[0]; i++) {
668 if (hugepg_tbl[i].orig_va) {
669 munmap(hugepg_tbl[i].orig_va, hpi->hugepage_sz);
670 hugepg_tbl[i].orig_va = NULL;
675 #endif /* RTE_EAL_SINGLE_FILE_SEGMENTS */
678 * Parse /proc/self/numa_maps to get the NUMA socket ID for each huge
682 find_numasocket(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi)
686 unsigned i, hp_count = 0;
689 char hugedir_str[PATH_MAX];
692 f = fopen("/proc/self/numa_maps", "r");
694 RTE_LOG(NOTICE, EAL, "cannot open /proc/self/numa_maps,"
695 " consider that all memory is in socket_id 0\n");
699 snprintf(hugedir_str, sizeof(hugedir_str),
700 "%s/%s", hpi->hugedir, internal_config.hugefile_prefix);
703 while (fgets(buf, sizeof(buf), f) != NULL) {
705 /* ignore non huge page */
706 if (strstr(buf, " huge ") == NULL &&
707 strstr(buf, hugedir_str) == NULL)
711 virt_addr = strtoull(buf, &end, 16);
712 if (virt_addr == 0 || end == buf) {
713 RTE_LOG(ERR, EAL, "%s(): error in numa_maps parsing\n", __func__);
717 /* get node id (socket id) */
718 nodestr = strstr(buf, " N");
719 if (nodestr == NULL) {
720 RTE_LOG(ERR, EAL, "%s(): error in numa_maps parsing\n", __func__);
724 end = strstr(nodestr, "=");
726 RTE_LOG(ERR, EAL, "%s(): error in numa_maps parsing\n", __func__);
732 socket_id = strtoul(nodestr, &end, 0);
733 if ((nodestr[0] == '\0') || (end == NULL) || (*end != '\0')) {
734 RTE_LOG(ERR, EAL, "%s(): error in numa_maps parsing\n", __func__);
738 /* if we find this page in our mappings, set socket_id */
739 for (i = 0; i < hpi->num_pages[0]; i++) {
740 void *va = (void *)(unsigned long)virt_addr;
741 if (hugepg_tbl[i].orig_va == va) {
742 hugepg_tbl[i].socket_id = socket_id;
748 if (hp_count < hpi->num_pages[0])
760 cmp_physaddr(const void *a, const void *b)
762 #ifndef RTE_ARCH_PPC_64
763 const struct hugepage_file *p1 = (const struct hugepage_file *)a;
764 const struct hugepage_file *p2 = (const struct hugepage_file *)b;
766 /* PowerPC needs memory sorted in reverse order from x86 */
767 const struct hugepage_file *p1 = (const struct hugepage_file *)b;
768 const struct hugepage_file *p2 = (const struct hugepage_file *)a;
770 if (p1->physaddr < p2->physaddr)
772 else if (p1->physaddr > p2->physaddr)
779 * Uses mmap to create a shared memory area for storage of data
780 * Used in this file to store the hugepage file map on disk
783 create_shared_memory(const char *filename, const size_t mem_size)
786 int fd = open(filename, O_CREAT | O_RDWR, 0666);
789 if (ftruncate(fd, mem_size) < 0) {
793 retval = mmap(NULL, mem_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
799 * this copies *active* hugepages from one hugepage table to another.
800 * destination is typically the shared memory.
803 copy_hugepages_to_shared_mem(struct hugepage_file * dst, int dest_size,
804 const struct hugepage_file * src, int src_size)
806 int src_pos, dst_pos = 0;
808 for (src_pos = 0; src_pos < src_size; src_pos++) {
809 if (src[src_pos].final_va != NULL) {
810 /* error on overflow attempt */
811 if (dst_pos == dest_size)
813 memcpy(&dst[dst_pos], &src[src_pos], sizeof(struct hugepage_file));
821 unlink_hugepage_files(struct hugepage_file *hugepg_tbl,
822 unsigned num_hp_info)
824 unsigned socket, size;
825 int page, nrpages = 0;
827 /* get total number of hugepages */
828 for (size = 0; size < num_hp_info; size++)
829 for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++)
831 internal_config.hugepage_info[size].num_pages[socket];
833 for (page = 0; page < nrpages; page++) {
834 struct hugepage_file *hp = &hugepg_tbl[page];
836 if (hp->final_va != NULL && unlink(hp->filepath)) {
837 RTE_LOG(WARNING, EAL, "%s(): Removing %s failed: %s\n",
838 __func__, hp->filepath, strerror(errno));
845 * unmaps hugepages that are not going to be used. since we originally allocate
846 * ALL hugepages (not just those we need), additional unmapping needs to be done.
849 unmap_unneeded_hugepages(struct hugepage_file *hugepg_tbl,
850 struct hugepage_info *hpi,
851 unsigned num_hp_info)
853 unsigned socket, size;
854 int page, nrpages = 0;
856 /* get total number of hugepages */
857 for (size = 0; size < num_hp_info; size++)
858 for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++)
859 nrpages += internal_config.hugepage_info[size].num_pages[socket];
861 for (size = 0; size < num_hp_info; size++) {
862 for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
863 unsigned pages_found = 0;
865 /* traverse until we have unmapped all the unused pages */
866 for (page = 0; page < nrpages; page++) {
867 struct hugepage_file *hp = &hugepg_tbl[page];
869 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
870 /* if this page was already cleared */
871 if (hp->final_va == NULL)
875 /* find a page that matches the criteria */
876 if ((hp->size == hpi[size].hugepage_sz) &&
877 (hp->socket_id == (int) socket)) {
879 /* if we skipped enough pages, unmap the rest */
880 if (pages_found == hpi[size].num_pages[socket]) {
883 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
884 unmap_len = hp->size * hp->repeated;
886 unmap_len = hp->size;
889 /* get start addr and len of the remaining segment */
890 munmap(hp->final_va, (size_t) unmap_len);
893 if (unlink(hp->filepath) == -1) {
894 RTE_LOG(ERR, EAL, "%s(): Removing %s failed: %s\n",
895 __func__, hp->filepath, strerror(errno));
899 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
900 /* else, check how much do we need to map */
903 hpi[size].num_pages[socket] - pages_found;
905 /* if we need enough memory to fit into the segment */
906 if (hp->repeated <= nr_pg_left) {
907 pages_found += hp->repeated;
909 /* truncate the segment */
911 uint64_t final_size = nr_pg_left * hp->size;
912 uint64_t seg_size = hp->repeated * hp->size;
914 void * unmap_va = RTE_PTR_ADD(hp->final_va,
918 munmap(unmap_va, seg_size - final_size);
920 fd = open(hp->filepath, O_RDWR);
922 RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
923 hp->filepath, strerror(errno));
926 if (ftruncate(fd, final_size) < 0) {
927 RTE_LOG(ERR, EAL, "Cannot truncate %s: %s\n",
928 hp->filepath, strerror(errno));
933 pages_found += nr_pg_left;
934 hp->repeated = nr_pg_left;
938 /* else, lock the page and skip */
945 } /* foreach socket */
946 } /* foreach pagesize */
951 static inline uint64_t
952 get_socket_mem_size(int socket)
957 for (i = 0; i < internal_config.num_hugepage_sizes; i++){
958 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
959 if (hpi->hugedir != NULL)
960 size += hpi->hugepage_sz * hpi->num_pages[socket];
967 * This function is a NUMA-aware equivalent of calc_num_pages.
968 * It takes in the list of hugepage sizes and the
969 * number of pages thereof, and calculates the best number of
970 * pages of each size to fulfill the request for <memory> ram
973 calc_num_pages_per_socket(uint64_t * memory,
974 struct hugepage_info *hp_info,
975 struct hugepage_info *hp_used,
976 unsigned num_hp_info)
978 unsigned socket, j, i = 0;
979 unsigned requested, available;
980 int total_num_pages = 0;
981 uint64_t remaining_mem, cur_mem;
982 uint64_t total_mem = internal_config.memory;
984 if (num_hp_info == 0)
987 /* if specific memory amounts per socket weren't requested */
988 if (internal_config.force_sockets == 0) {
989 int cpu_per_socket[RTE_MAX_NUMA_NODES];
990 size_t default_size, total_size;
993 /* Compute number of cores per socket */
994 memset(cpu_per_socket, 0, sizeof(cpu_per_socket));
995 RTE_LCORE_FOREACH(lcore_id) {
996 cpu_per_socket[rte_lcore_to_socket_id(lcore_id)]++;
1000 * Automatically spread requested memory amongst detected sockets according
1001 * to number of cores from cpu mask present on each socket
1003 total_size = internal_config.memory;
1004 for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_size != 0; socket++) {
1006 /* Set memory amount per socket */
1007 default_size = (internal_config.memory * cpu_per_socket[socket])
1008 / rte_lcore_count();
1010 /* Limit to maximum available memory on socket */
1011 default_size = RTE_MIN(default_size, get_socket_mem_size(socket));
1014 memory[socket] = default_size;
1015 total_size -= default_size;
1019 * If some memory is remaining, try to allocate it by getting all
1020 * available memory from sockets, one after the other
1022 for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_size != 0; socket++) {
1023 /* take whatever is available */
1024 default_size = RTE_MIN(get_socket_mem_size(socket) - memory[socket],
1028 memory[socket] += default_size;
1029 total_size -= default_size;
1033 for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_mem != 0; socket++) {
1034 /* skips if the memory on specific socket wasn't requested */
1035 for (i = 0; i < num_hp_info && memory[socket] != 0; i++){
1036 hp_used[i].hugedir = hp_info[i].hugedir;
1037 hp_used[i].num_pages[socket] = RTE_MIN(
1038 memory[socket] / hp_info[i].hugepage_sz,
1039 hp_info[i].num_pages[socket]);
1041 cur_mem = hp_used[i].num_pages[socket] *
1042 hp_used[i].hugepage_sz;
1044 memory[socket] -= cur_mem;
1045 total_mem -= cur_mem;
1047 total_num_pages += hp_used[i].num_pages[socket];
1049 /* check if we have met all memory requests */
1050 if (memory[socket] == 0)
1053 /* check if we have any more pages left at this size, if so
1054 * move on to next size */
1055 if (hp_used[i].num_pages[socket] == hp_info[i].num_pages[socket])
1057 /* At this point we know that there are more pages available that are
1058 * bigger than the memory we want, so lets see if we can get enough
1059 * from other page sizes.
1062 for (j = i+1; j < num_hp_info; j++)
1063 remaining_mem += hp_info[j].hugepage_sz *
1064 hp_info[j].num_pages[socket];
1066 /* is there enough other memory, if not allocate another page and quit */
1067 if (remaining_mem < memory[socket]){
1068 cur_mem = RTE_MIN(memory[socket],
1069 hp_info[i].hugepage_sz);
1070 memory[socket] -= cur_mem;
1071 total_mem -= cur_mem;
1072 hp_used[i].num_pages[socket]++;
1074 break; /* we are done with this socket*/
1077 /* if we didn't satisfy all memory requirements per socket */
1078 if (memory[socket] > 0) {
1079 /* to prevent icc errors */
1080 requested = (unsigned) (internal_config.socket_mem[socket] /
1082 available = requested -
1083 ((unsigned) (memory[socket] / 0x100000));
1084 RTE_LOG(ERR, EAL, "Not enough memory available on socket %u! "
1085 "Requested: %uMB, available: %uMB\n", socket,
1086 requested, available);
1091 /* if we didn't satisfy total memory requirements */
1092 if (total_mem > 0) {
1093 requested = (unsigned) (internal_config.memory / 0x100000);
1094 available = requested - (unsigned) (total_mem / 0x100000);
1095 RTE_LOG(ERR, EAL, "Not enough memory available! Requested: %uMB,"
1096 " available: %uMB\n", requested, available);
1099 return total_num_pages;
1102 static inline size_t
1103 eal_get_hugepage_mem_size(void)
1108 for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
1109 struct hugepage_info *hpi = &internal_config.hugepage_info[i];
1110 if (hpi->hugedir != NULL) {
1111 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
1112 size += hpi->hugepage_sz * hpi->num_pages[j];
1117 return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
1120 static struct sigaction huge_action_old;
1121 static int huge_need_recover;
1124 huge_register_sigbus(void)
1127 struct sigaction action;
1130 sigaddset(&mask, SIGBUS);
1131 action.sa_flags = 0;
1132 action.sa_mask = mask;
1133 action.sa_handler = huge_sigbus_handler;
1135 huge_need_recover = !sigaction(SIGBUS, &action, &huge_action_old);
1139 huge_recover_sigbus(void)
1141 if (huge_need_recover) {
1142 sigaction(SIGBUS, &huge_action_old, NULL);
1143 huge_need_recover = 0;
1148 * Prepare physical memory mapping: fill configuration structure with
1149 * these infos, return 0 on success.
1150 * 1. map N huge pages in separate files in hugetlbfs
1151 * 2. find associated physical addr
1152 * 3. find associated NUMA socket ID
1153 * 4. sort all huge pages by physical address
1154 * 5. remap these N huge pages in the correct order
1155 * 6. unmap the first mapping
1156 * 7. fill memsegs in configuration with contiguous zones
1159 rte_eal_hugepage_init(void)
1161 struct rte_mem_config *mcfg;
1162 struct hugepage_file *hugepage = NULL, *tmp_hp = NULL;
1163 struct hugepage_info used_hp[MAX_HUGEPAGE_SIZES];
1165 uint64_t memory[RTE_MAX_NUMA_NODES];
1168 int i, j, new_memseg;
1169 int nr_hugefiles, nr_hugepages = 0;
1171 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1172 int new_pages_count[MAX_HUGEPAGE_SIZES];
1175 test_proc_pagemap_readable();
1177 memset(used_hp, 0, sizeof(used_hp));
1179 /* get pointer to global configuration */
1180 mcfg = rte_eal_get_configuration()->mem_config;
1182 /* hugetlbfs can be disabled */
1183 if (internal_config.no_hugetlbfs) {
1184 addr = mmap(NULL, internal_config.memory, PROT_READ | PROT_WRITE,
1185 MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
1186 if (addr == MAP_FAILED) {
1187 RTE_LOG(ERR, EAL, "%s: mmap() failed: %s\n", __func__,
1191 mcfg->memseg[0].phys_addr = (phys_addr_t)(uintptr_t)addr;
1192 mcfg->memseg[0].addr = addr;
1193 mcfg->memseg[0].hugepage_sz = RTE_PGSIZE_4K;
1194 mcfg->memseg[0].len = internal_config.memory;
1195 mcfg->memseg[0].socket_id = 0;
1199 /* check if app runs on Xen Dom0 */
1200 if (internal_config.xen_dom0_support) {
1201 #ifdef RTE_LIBRTE_XEN_DOM0
1202 /* use dom0_mm kernel driver to init memory */
1203 if (rte_xen_dom0_memory_init() < 0)
1210 /* calculate total number of hugepages available. at this point we haven't
1211 * yet started sorting them so they all are on socket 0 */
1212 for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++) {
1213 /* meanwhile, also initialize used_hp hugepage sizes in used_hp */
1214 used_hp[i].hugepage_sz = internal_config.hugepage_info[i].hugepage_sz;
1216 nr_hugepages += internal_config.hugepage_info[i].num_pages[0];
1220 * allocate a memory area for hugepage table.
1221 * this isn't shared memory yet. due to the fact that we need some
1222 * processing done on these pages, shared memory will be created
1225 tmp_hp = malloc(nr_hugepages * sizeof(struct hugepage_file));
1229 memset(tmp_hp, 0, nr_hugepages * sizeof(struct hugepage_file));
1231 hp_offset = 0; /* where we start the current page size entries */
1233 huge_register_sigbus();
1235 /* map all hugepages and sort them */
1236 for (i = 0; i < (int)internal_config.num_hugepage_sizes; i ++){
1237 unsigned pages_old, pages_new;
1238 struct hugepage_info *hpi;
1241 * we don't yet mark hugepages as used at this stage, so
1242 * we just map all hugepages available to the system
1243 * all hugepages are still located on socket 0
1245 hpi = &internal_config.hugepage_info[i];
1247 if (hpi->num_pages[0] == 0)
1250 /* map all hugepages available */
1251 pages_old = hpi->num_pages[0];
1252 pages_new = map_all_hugepages(&tmp_hp[hp_offset], hpi, 1);
1253 if (pages_new < pages_old) {
1254 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1256 "%d not %d hugepages of size %u MB allocated\n",
1257 pages_new, pages_old,
1258 (unsigned)(hpi->hugepage_sz / 0x100000));
1262 "%d not %d hugepages of size %u MB allocated\n",
1263 pages_new, pages_old,
1264 (unsigned)(hpi->hugepage_sz / 0x100000));
1266 int pages = pages_old - pages_new;
1268 nr_hugepages -= pages;
1269 hpi->num_pages[0] = pages_new;
1275 /* find physical addresses and sockets for each hugepage */
1276 if (find_physaddrs(&tmp_hp[hp_offset], hpi) < 0){
1277 RTE_LOG(DEBUG, EAL, "Failed to find phys addr for %u MB pages\n",
1278 (unsigned)(hpi->hugepage_sz / 0x100000));
1282 if (find_numasocket(&tmp_hp[hp_offset], hpi) < 0){
1283 RTE_LOG(DEBUG, EAL, "Failed to find NUMA socket for %u MB pages\n",
1284 (unsigned)(hpi->hugepage_sz / 0x100000));
1288 qsort(&tmp_hp[hp_offset], hpi->num_pages[0],
1289 sizeof(struct hugepage_file), cmp_physaddr);
1291 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1292 /* remap all hugepages into single file segments */
1293 new_pages_count[i] = remap_all_hugepages(&tmp_hp[hp_offset], hpi);
1294 if (new_pages_count[i] < 0){
1295 RTE_LOG(DEBUG, EAL, "Failed to remap %u MB pages\n",
1296 (unsigned)(hpi->hugepage_sz / 0x100000));
1300 /* we have processed a num of hugepages of this size, so inc offset */
1301 hp_offset += new_pages_count[i];
1303 /* remap all hugepages */
1304 if (map_all_hugepages(&tmp_hp[hp_offset], hpi, 0) !=
1305 hpi->num_pages[0]) {
1306 RTE_LOG(ERR, EAL, "Failed to remap %u MB pages\n",
1307 (unsigned)(hpi->hugepage_sz / 0x100000));
1311 /* unmap original mappings */
1312 if (unmap_all_hugepages_orig(&tmp_hp[hp_offset], hpi) < 0)
1315 /* we have processed a num of hugepages of this size, so inc offset */
1316 hp_offset += hpi->num_pages[0];
1320 huge_recover_sigbus();
1322 if (internal_config.memory == 0 && internal_config.force_sockets == 0)
1323 internal_config.memory = eal_get_hugepage_mem_size();
1325 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1327 for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++) {
1328 nr_hugefiles += new_pages_count[i];
1331 nr_hugefiles = nr_hugepages;
1335 /* clean out the numbers of pages */
1336 for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++)
1337 for (j = 0; j < RTE_MAX_NUMA_NODES; j++)
1338 internal_config.hugepage_info[i].num_pages[j] = 0;
1340 /* get hugepages for each socket */
1341 for (i = 0; i < nr_hugefiles; i++) {
1342 int socket = tmp_hp[i].socket_id;
1344 /* find a hugepage info with right size and increment num_pages */
1345 const int nb_hpsizes = RTE_MIN(MAX_HUGEPAGE_SIZES,
1346 (int)internal_config.num_hugepage_sizes);
1347 for (j = 0; j < nb_hpsizes; j++) {
1348 if (tmp_hp[i].size ==
1349 internal_config.hugepage_info[j].hugepage_sz) {
1350 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1351 internal_config.hugepage_info[j].num_pages[socket] +=
1354 internal_config.hugepage_info[j].num_pages[socket]++;
1360 /* make a copy of socket_mem, needed for number of pages calculation */
1361 for (i = 0; i < RTE_MAX_NUMA_NODES; i++)
1362 memory[i] = internal_config.socket_mem[i];
1364 /* calculate final number of pages */
1365 nr_hugepages = calc_num_pages_per_socket(memory,
1366 internal_config.hugepage_info, used_hp,
1367 internal_config.num_hugepage_sizes);
1369 /* error if not enough memory available */
1370 if (nr_hugepages < 0)
1374 for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++) {
1375 for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
1376 if (used_hp[i].num_pages[j] > 0) {
1378 "Requesting %u pages of size %uMB"
1379 " from socket %i\n",
1380 used_hp[i].num_pages[j],
1382 (used_hp[i].hugepage_sz / 0x100000),
1388 /* create shared memory */
1389 hugepage = create_shared_memory(eal_hugepage_info_path(),
1390 nr_hugefiles * sizeof(struct hugepage_file));
1392 if (hugepage == NULL) {
1393 RTE_LOG(ERR, EAL, "Failed to create shared memory!\n");
1396 memset(hugepage, 0, nr_hugefiles * sizeof(struct hugepage_file));
1399 * unmap pages that we won't need (looks at used_hp).
1400 * also, sets final_va to NULL on pages that were unmapped.
1402 if (unmap_unneeded_hugepages(tmp_hp, used_hp,
1403 internal_config.num_hugepage_sizes) < 0) {
1404 RTE_LOG(ERR, EAL, "Unmapping and locking hugepages failed!\n");
1409 * copy stuff from malloc'd hugepage* to the actual shared memory.
1410 * this procedure only copies those hugepages that have final_va
1411 * not NULL. has overflow protection.
1413 if (copy_hugepages_to_shared_mem(hugepage, nr_hugefiles,
1414 tmp_hp, nr_hugefiles) < 0) {
1415 RTE_LOG(ERR, EAL, "Copying tables to shared memory failed!\n");
1419 /* free the hugepage backing files */
1420 if (internal_config.hugepage_unlink &&
1421 unlink_hugepage_files(tmp_hp, internal_config.num_hugepage_sizes) < 0) {
1422 RTE_LOG(ERR, EAL, "Unlinking hugepage files failed!\n");
1426 /* free the temporary hugepage table */
1430 /* find earliest free memseg - this is needed because in case of IVSHMEM,
1431 * segments might have already been initialized */
1432 for (j = 0; j < RTE_MAX_MEMSEG; j++)
1433 if (mcfg->memseg[j].addr == NULL) {
1434 /* move to previous segment and exit loop */
1439 for (i = 0; i < nr_hugefiles; i++) {
1442 /* if this is a new section, create a new memseg */
1445 else if (hugepage[i].socket_id != hugepage[i-1].socket_id)
1447 else if (hugepage[i].size != hugepage[i-1].size)
1450 #ifdef RTE_ARCH_PPC_64
1451 /* On PPC64 architecture, the mmap always start from higher
1452 * virtual address to lower address. Here, both the physical
1453 * address and virtual address are in descending order */
1454 else if ((hugepage[i-1].physaddr - hugepage[i].physaddr) !=
1457 else if (((unsigned long)hugepage[i-1].final_va -
1458 (unsigned long)hugepage[i].final_va) != hugepage[i].size)
1461 else if ((hugepage[i].physaddr - hugepage[i-1].physaddr) !=
1464 else if (((unsigned long)hugepage[i].final_va -
1465 (unsigned long)hugepage[i-1].final_va) != hugepage[i].size)
1471 if (j == RTE_MAX_MEMSEG)
1474 mcfg->memseg[j].phys_addr = hugepage[i].physaddr;
1475 mcfg->memseg[j].addr = hugepage[i].final_va;
1476 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1477 mcfg->memseg[j].len = hugepage[i].size * hugepage[i].repeated;
1479 mcfg->memseg[j].len = hugepage[i].size;
1481 mcfg->memseg[j].socket_id = hugepage[i].socket_id;
1482 mcfg->memseg[j].hugepage_sz = hugepage[i].size;
1484 /* continuation of previous memseg */
1486 #ifdef RTE_ARCH_PPC_64
1487 /* Use the phy and virt address of the last page as segment
1488 * address for IBM Power architecture */
1489 mcfg->memseg[j].phys_addr = hugepage[i].physaddr;
1490 mcfg->memseg[j].addr = hugepage[i].final_va;
1492 mcfg->memseg[j].len += mcfg->memseg[j].hugepage_sz;
1494 hugepage[i].memseg_id = j;
1497 if (i < nr_hugefiles) {
1498 RTE_LOG(ERR, EAL, "Can only reserve %d pages "
1499 "from %d requested\n"
1500 "Current %s=%d is not enough\n"
1501 "Please either increase it or request less amount "
1503 i, nr_hugefiles, RTE_STR(CONFIG_RTE_MAX_MEMSEG),
1508 munmap(hugepage, nr_hugefiles * sizeof(struct hugepage_file));
1513 huge_recover_sigbus();
1515 if (hugepage != NULL)
1516 munmap(hugepage, nr_hugefiles * sizeof(struct hugepage_file));
1522 * uses fstat to report the size of a file on disk
1528 if (fstat(fd, &st) < 0)
1534 * This creates the memory mappings in the secondary process to match that of
1535 * the server process. It goes through each memory segment in the DPDK runtime
1536 * configuration and finds the hugepages which form that segment, mapping them
1537 * in order to form a contiguous block in the virtual memory space
1540 rte_eal_hugepage_attach(void)
1542 const struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
1543 struct hugepage_file *hp = NULL;
1544 unsigned num_hp = 0;
1545 unsigned i, s = 0; /* s used to track the segment number */
1547 int fd, fd_zero = -1, fd_hugepage = -1;
1549 if (aslr_enabled() > 0) {
1550 RTE_LOG(WARNING, EAL, "WARNING: Address Space Layout Randomization "
1551 "(ASLR) is enabled in the kernel.\n");
1552 RTE_LOG(WARNING, EAL, " This may cause issues with mapping memory "
1553 "into secondary processes\n");
1556 test_proc_pagemap_readable();
1558 if (internal_config.xen_dom0_support) {
1559 #ifdef RTE_LIBRTE_XEN_DOM0
1560 if (rte_xen_dom0_memory_attach() < 0) {
1561 RTE_LOG(ERR, EAL, "Failed to attach memory segments of primary "
1569 fd_zero = open("/dev/zero", O_RDONLY);
1571 RTE_LOG(ERR, EAL, "Could not open /dev/zero\n");
1574 fd_hugepage = open(eal_hugepage_info_path(), O_RDONLY);
1575 if (fd_hugepage < 0) {
1576 RTE_LOG(ERR, EAL, "Could not open %s\n", eal_hugepage_info_path());
1580 /* map all segments into memory to make sure we get the addrs */
1581 for (s = 0; s < RTE_MAX_MEMSEG; ++s) {
1585 * the first memory segment with len==0 is the one that
1586 * follows the last valid segment.
1588 if (mcfg->memseg[s].len == 0)
1591 #ifdef RTE_LIBRTE_IVSHMEM
1593 * if segment has ioremap address set, it's an IVSHMEM segment and
1594 * doesn't need mapping as it was already mapped earlier
1596 if (mcfg->memseg[s].ioremap_addr != 0)
1601 * fdzero is mmapped to get a contiguous block of virtual
1602 * addresses of the appropriate memseg size.
1603 * use mmap to get identical addresses as the primary process.
1605 base_addr = mmap(mcfg->memseg[s].addr, mcfg->memseg[s].len,
1606 PROT_READ, MAP_PRIVATE, fd_zero, 0);
1607 if (base_addr == MAP_FAILED ||
1608 base_addr != mcfg->memseg[s].addr) {
1609 RTE_LOG(ERR, EAL, "Could not mmap %llu bytes "
1610 "in /dev/zero to requested address [%p]: '%s'\n",
1611 (unsigned long long)mcfg->memseg[s].len,
1612 mcfg->memseg[s].addr, strerror(errno));
1613 if (aslr_enabled() > 0) {
1614 RTE_LOG(ERR, EAL, "It is recommended to "
1615 "disable ASLR in the kernel "
1616 "and retry running both primary "
1617 "and secondary processes\n");
1623 size = getFileSize(fd_hugepage);
1624 hp = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd_hugepage, 0);
1625 if (hp == MAP_FAILED) {
1626 RTE_LOG(ERR, EAL, "Could not mmap %s\n", eal_hugepage_info_path());
1630 num_hp = size / sizeof(struct hugepage_file);
1631 RTE_LOG(DEBUG, EAL, "Analysing %u files\n", num_hp);
1634 while (s < RTE_MAX_MEMSEG && mcfg->memseg[s].len > 0){
1635 void *addr, *base_addr;
1636 uintptr_t offset = 0;
1637 size_t mapping_size;
1638 #ifdef RTE_LIBRTE_IVSHMEM
1640 * if segment has ioremap address set, it's an IVSHMEM segment and
1641 * doesn't need mapping as it was already mapped earlier
1643 if (mcfg->memseg[s].ioremap_addr != 0) {
1649 * free previously mapped memory so we can map the
1650 * hugepages into the space
1652 base_addr = mcfg->memseg[s].addr;
1653 munmap(base_addr, mcfg->memseg[s].len);
1655 /* find the hugepages for this segment and map them
1656 * we don't need to worry about order, as the server sorted the
1657 * entries before it did the second mmap of them */
1658 for (i = 0; i < num_hp && offset < mcfg->memseg[s].len; i++){
1659 if (hp[i].memseg_id == (int)s){
1660 fd = open(hp[i].filepath, O_RDWR);
1662 RTE_LOG(ERR, EAL, "Could not open %s\n",
1666 #ifdef RTE_EAL_SINGLE_FILE_SEGMENTS
1667 mapping_size = hp[i].size * hp[i].repeated;
1669 mapping_size = hp[i].size;
1671 addr = mmap(RTE_PTR_ADD(base_addr, offset),
1672 mapping_size, PROT_READ | PROT_WRITE,
1674 close(fd); /* close file both on success and on failure */
1675 if (addr == MAP_FAILED ||
1676 addr != RTE_PTR_ADD(base_addr, offset)) {
1677 RTE_LOG(ERR, EAL, "Could not mmap %s\n",
1681 offset+=mapping_size;
1684 RTE_LOG(DEBUG, EAL, "Mapped segment %u of size 0x%llx\n", s,
1685 (unsigned long long)mcfg->memseg[s].len);
1688 /* unmap the hugepage config file, since we are done using it */
1696 while (s < RTE_MAX_MEMSEG && mcfg->memseg[s].len > 0) {
1697 munmap(mcfg->memseg[s].addr, mcfg->memseg[s].len);
1700 if (hp != NULL && hp != MAP_FAILED)
1704 if (fd_hugepage >= 0)