diff options
Diffstat (limited to 'lib/librte_eal/linuxapp/eal/eal_memory.c')
| -rw-r--r-- | lib/librte_eal/linuxapp/eal/eal_memory.c | 1520 |
1 files changed, 1123 insertions, 397 deletions
diff --git a/lib/librte_eal/linuxapp/eal/eal_memory.c b/lib/librte_eal/linuxapp/eal/eal_memory.c index 38853b75..dbf19499 100644 --- a/lib/librte_eal/linuxapp/eal/eal_memory.c +++ b/lib/librte_eal/linuxapp/eal/eal_memory.c @@ -28,6 +28,7 @@ #include <numaif.h> #endif +#include <rte_errno.h> #include <rte_log.h> #include <rte_memory.h> #include <rte_launch.h> @@ -39,6 +40,7 @@ #include <rte_string_fns.h> #include "eal_private.h" +#include "eal_memalloc.h" #include "eal_internal_cfg.h" #include "eal_filesystem.h" #include "eal_hugepages.h" @@ -57,8 +59,6 @@ * zone as well as a physical contiguous zone. */ -static uint64_t baseaddr_offset; - static bool phys_addrs_available = true; #define RANDOMIZE_VA_SPACE_FILE "/proc/sys/kernel/randomize_va_space" @@ -66,7 +66,7 @@ static bool phys_addrs_available = true; static void test_phys_addrs_available(void) { - uint64_t tmp; + uint64_t tmp = 0; phys_addr_t physaddr; if (!rte_eal_has_hugepages()) { @@ -221,82 +221,6 @@ aslr_enabled(void) } } -/* - * Try to mmap *size bytes in /dev/zero. If it is successful, return the - * pointer to the mmap'd area and keep *size unmodified. Else, retry - * with a smaller zone: decrease *size by hugepage_sz until it reaches - * 0. In this case, return NULL. Note: this function returns an address - * which is a multiple of hugepage size. - */ -static void * -get_virtual_area(size_t *size, size_t hugepage_sz) -{ - void *addr; - void *addr_hint; - int fd; - long aligned_addr; - - if (internal_config.base_virtaddr != 0) { - int page_size = sysconf(_SC_PAGE_SIZE); - addr_hint = (void *) (uintptr_t) - (internal_config.base_virtaddr + baseaddr_offset); - addr_hint = RTE_PTR_ALIGN_FLOOR(addr_hint, page_size); - } else { - addr_hint = NULL; - } - - RTE_LOG(DEBUG, EAL, "Ask a virtual area of 0x%zx bytes\n", *size); - - - fd = open("/dev/zero", O_RDONLY); - if (fd < 0){ - RTE_LOG(ERR, EAL, "Cannot open /dev/zero\n"); - return NULL; - } - do { - addr = mmap(addr_hint, (*size) + hugepage_sz, PROT_READ, -#ifdef RTE_ARCH_PPC_64 - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -#else - MAP_PRIVATE, -#endif - fd, 0); - if (addr == MAP_FAILED) { - *size -= hugepage_sz; - } else if (addr_hint != NULL && addr != addr_hint) { - RTE_LOG(WARNING, EAL, "WARNING! Base virtual address " - "hint (%p != %p) not respected!\n", - addr_hint, addr); - RTE_LOG(WARNING, EAL, " This may cause issues with " - "mapping memory into secondary processes\n"); - } - } while (addr == MAP_FAILED && *size > 0); - - if (addr == MAP_FAILED) { - close(fd); - RTE_LOG(ERR, EAL, "Cannot get a virtual area: %s\n", - strerror(errno)); - return NULL; - } - - munmap(addr, (*size) + hugepage_sz); - close(fd); - - /* align addr to a huge page size boundary */ - aligned_addr = (long)addr; - aligned_addr += (hugepage_sz - 1); - aligned_addr &= (~(hugepage_sz - 1)); - addr = (void *)(aligned_addr); - - RTE_LOG(DEBUG, EAL, "Virtual area found at %p (size = 0x%zx)\n", - addr, *size); - - /* increment offset */ - baseaddr_offset += *size; - - return addr; -} - static sigjmp_buf huge_jmpenv; static void huge_sigbus_handler(int signo __rte_unused) @@ -330,13 +254,11 @@ void numa_error(char *where) */ static unsigned map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, - uint64_t *essential_memory __rte_unused, int orig) + uint64_t *essential_memory __rte_unused) { int fd; unsigned i; void *virtaddr; - void *vma_addr = NULL; - size_t vma_len = 0; #ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES int node_id = -1; int essential_prev = 0; @@ -351,7 +273,7 @@ map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, have_numa = false; } - if (orig && have_numa) { + if (have_numa) { RTE_LOG(DEBUG, EAL, "Trying to obtain current memory policy.\n"); if (get_mempolicy(&oldpolicy, oldmask->maskp, oldmask->size + 1, 0, 0) < 0) { @@ -367,6 +289,7 @@ map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, #endif for (i = 0; i < hpi->num_pages[0]; i++) { + struct hugepage_file *hf = &hugepg_tbl[i]; uint64_t hugepage_sz = hpi->hugepage_sz; #ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES @@ -401,57 +324,14 @@ map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, } #endif - if (orig) { - hugepg_tbl[i].file_id = i; - hugepg_tbl[i].size = hugepage_sz; - eal_get_hugefile_path(hugepg_tbl[i].filepath, - sizeof(hugepg_tbl[i].filepath), hpi->hugedir, - hugepg_tbl[i].file_id); - hugepg_tbl[i].filepath[sizeof(hugepg_tbl[i].filepath) - 1] = '\0'; - } -#ifndef RTE_ARCH_64 - /* for 32-bit systems, don't remap 1G and 16G pages, just reuse - * original map address as final map address. - */ - else if ((hugepage_sz == RTE_PGSIZE_1G) - || (hugepage_sz == RTE_PGSIZE_16G)) { - hugepg_tbl[i].final_va = hugepg_tbl[i].orig_va; - hugepg_tbl[i].orig_va = NULL; - continue; - } -#endif - else if (vma_len == 0) { - unsigned j, num_pages; - - /* reserve a virtual area for next contiguous - * physical block: count the number of - * contiguous physical pages. */ - for (j = i+1; j < hpi->num_pages[0] ; j++) { -#ifdef RTE_ARCH_PPC_64 - /* The physical addresses are sorted in - * descending order on PPC64 */ - if (hugepg_tbl[j].physaddr != - hugepg_tbl[j-1].physaddr - hugepage_sz) - break; -#else - if (hugepg_tbl[j].physaddr != - hugepg_tbl[j-1].physaddr + hugepage_sz) - break; -#endif - } - num_pages = j - i; - vma_len = num_pages * hugepage_sz; - - /* get the biggest virtual memory area up to - * vma_len. If it fails, vma_addr is NULL, so - * let the kernel provide the address. */ - vma_addr = get_virtual_area(&vma_len, hpi->hugepage_sz); - if (vma_addr == NULL) - vma_len = hugepage_sz; - } + hf->file_id = i; + hf->size = hugepage_sz; + eal_get_hugefile_path(hf->filepath, sizeof(hf->filepath), + hpi->hugedir, hf->file_id); + hf->filepath[sizeof(hf->filepath) - 1] = '\0'; /* try to create hugepage file */ - fd = open(hugepg_tbl[i].filepath, O_CREAT | O_RDWR, 0600); + fd = open(hf->filepath, O_CREAT | O_RDWR, 0600); if (fd < 0) { RTE_LOG(DEBUG, EAL, "%s(): open failed: %s\n", __func__, strerror(errno)); @@ -459,8 +339,11 @@ map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, } /* map the segment, and populate page tables, - * the kernel fills this segment with zeros */ - virtaddr = mmap(vma_addr, hugepage_sz, PROT_READ | PROT_WRITE, + * the kernel fills this segment with zeros. we don't care where + * this gets mapped - we already have contiguous memory areas + * ready for us to map into. + */ + virtaddr = mmap(NULL, hugepage_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, 0); if (virtaddr == MAP_FAILED) { RTE_LOG(DEBUG, EAL, "%s(): mmap failed: %s\n", __func__, @@ -469,41 +352,33 @@ map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, goto out; } - if (orig) { - hugepg_tbl[i].orig_va = virtaddr; - } - else { - hugepg_tbl[i].final_va = virtaddr; - } + hf->orig_va = virtaddr; - if (orig) { - /* In linux, hugetlb limitations, like cgroup, are - * enforced at fault time instead of mmap(), even - * with the option of MAP_POPULATE. Kernel will send - * a SIGBUS signal. To avoid to be killed, save stack - * environment here, if SIGBUS happens, we can jump - * back here. - */ - if (huge_wrap_sigsetjmp()) { - RTE_LOG(DEBUG, EAL, "SIGBUS: Cannot mmap more " - "hugepages of size %u MB\n", - (unsigned)(hugepage_sz / 0x100000)); - munmap(virtaddr, hugepage_sz); - close(fd); - unlink(hugepg_tbl[i].filepath); + /* In linux, hugetlb limitations, like cgroup, are + * enforced at fault time instead of mmap(), even + * with the option of MAP_POPULATE. Kernel will send + * a SIGBUS signal. To avoid to be killed, save stack + * environment here, if SIGBUS happens, we can jump + * back here. + */ + if (huge_wrap_sigsetjmp()) { + RTE_LOG(DEBUG, EAL, "SIGBUS: Cannot mmap more " + "hugepages of size %u MB\n", + (unsigned int)(hugepage_sz / 0x100000)); + munmap(virtaddr, hugepage_sz); + close(fd); + unlink(hugepg_tbl[i].filepath); #ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES - if (maxnode) - essential_memory[node_id] = - essential_prev; + if (maxnode) + essential_memory[node_id] = + essential_prev; #endif - goto out; - } - *(int *)virtaddr = 0; + goto out; } + *(int *)virtaddr = 0; - - /* set shared flock on the file. */ - if (flock(fd, LOCK_SH | LOCK_NB) == -1) { + /* set shared lock on the file. */ + if (flock(fd, LOCK_SH) < 0) { RTE_LOG(DEBUG, EAL, "%s(): Locking file failed:%s \n", __func__, strerror(errno)); close(fd); @@ -511,9 +386,6 @@ map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, } close(fd); - - vma_addr = (char *)vma_addr + hugepage_sz; - vma_len -= hugepage_sz; } out: @@ -535,20 +407,6 @@ out: return i; } -/* Unmap all hugepages from original mapping */ -static int -unmap_all_hugepages_orig(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi) -{ - unsigned i; - for (i = 0; i < hpi->num_pages[0]; i++) { - if (hugepg_tbl[i].orig_va) { - munmap(hugepg_tbl[i].orig_va, hpi->hugepage_sz); - hugepg_tbl[i].orig_va = NULL; - } - } - return 0; -} - /* * Parse /proc/self/numa_maps to get the NUMA socket ID for each huge * page. @@ -663,7 +521,18 @@ static void * create_shared_memory(const char *filename, const size_t mem_size) { void *retval; - int fd = open(filename, O_CREAT | O_RDWR, 0666); + int fd; + + /* if no shared files mode is used, create anonymous memory instead */ + if (internal_config.no_shconf) { + retval = mmap(NULL, mem_size, PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + if (retval == MAP_FAILED) + return NULL; + return retval; + } + + fd = open(filename, O_CREAT | O_RDWR, 0666); if (fd < 0) return NULL; if (ftruncate(fd, mem_size) < 0) { @@ -688,7 +557,7 @@ copy_hugepages_to_shared_mem(struct hugepage_file * dst, int dest_size, int src_pos, dst_pos = 0; for (src_pos = 0; src_pos < src_size; src_pos++) { - if (src[src_pos].final_va != NULL) { + if (src[src_pos].orig_va != NULL) { /* error on overflow attempt */ if (dst_pos == dest_size) return -1; @@ -759,9 +628,10 @@ unmap_unneeded_hugepages(struct hugepage_file *hugepg_tbl, unmap_len = hp->size; /* get start addr and len of the remaining segment */ - munmap(hp->final_va, (size_t) unmap_len); + munmap(hp->orig_va, + (size_t)unmap_len); - hp->final_va = NULL; + hp->orig_va = NULL; if (unlink(hp->filepath) == -1) { RTE_LOG(ERR, EAL, "%s(): Removing %s failed: %s\n", __func__, hp->filepath, strerror(errno)); @@ -780,6 +650,436 @@ unmap_unneeded_hugepages(struct hugepage_file *hugepg_tbl, return 0; } +static int +remap_segment(struct hugepage_file *hugepages, int seg_start, int seg_end) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + struct rte_memseg_list *msl; + struct rte_fbarray *arr; + int cur_page, seg_len; + unsigned int msl_idx; + int ms_idx; + uint64_t page_sz; + size_t memseg_len; + int socket_id; + + page_sz = hugepages[seg_start].size; + socket_id = hugepages[seg_start].socket_id; + seg_len = seg_end - seg_start; + + RTE_LOG(DEBUG, EAL, "Attempting to map %" PRIu64 "M on socket %i\n", + (seg_len * page_sz) >> 20ULL, socket_id); + + /* find free space in memseg lists */ + for (msl_idx = 0; msl_idx < RTE_MAX_MEMSEG_LISTS; msl_idx++) { + bool empty; + msl = &mcfg->memsegs[msl_idx]; + arr = &msl->memseg_arr; + + if (msl->page_sz != page_sz) + continue; + if (msl->socket_id != socket_id) + continue; + + /* leave space for a hole if array is not empty */ + empty = arr->count == 0; + ms_idx = rte_fbarray_find_next_n_free(arr, 0, + seg_len + (empty ? 0 : 1)); + + /* memseg list is full? */ + if (ms_idx < 0) + continue; + + /* leave some space between memsegs, they are not IOVA + * contiguous, so they shouldn't be VA contiguous either. + */ + if (!empty) + ms_idx++; + break; + } + if (msl_idx == RTE_MAX_MEMSEG_LISTS) { + RTE_LOG(ERR, EAL, "Could not find space for memseg. Please increase %s and/or %s in configuration.\n", + RTE_STR(CONFIG_RTE_MAX_MEMSEG_PER_TYPE), + RTE_STR(CONFIG_RTE_MAX_MEM_PER_TYPE)); + return -1; + } + +#ifdef RTE_ARCH_PPC64 + /* for PPC64 we go through the list backwards */ + for (cur_page = seg_end - 1; cur_page >= seg_start; + cur_page--, ms_idx++) { +#else + for (cur_page = seg_start; cur_page < seg_end; cur_page++, ms_idx++) { +#endif + struct hugepage_file *hfile = &hugepages[cur_page]; + struct rte_memseg *ms = rte_fbarray_get(arr, ms_idx); + void *addr; + int fd; + + fd = open(hfile->filepath, O_RDWR); + if (fd < 0) { + RTE_LOG(ERR, EAL, "Could not open '%s': %s\n", + hfile->filepath, strerror(errno)); + return -1; + } + /* set shared lock on the file. */ + if (flock(fd, LOCK_SH) < 0) { + RTE_LOG(DEBUG, EAL, "Could not lock '%s': %s\n", + hfile->filepath, strerror(errno)); + close(fd); + return -1; + } + memseg_len = (size_t)page_sz; + addr = RTE_PTR_ADD(msl->base_va, ms_idx * memseg_len); + + /* we know this address is already mmapped by memseg list, so + * using MAP_FIXED here is safe + */ + addr = mmap(addr, page_sz, PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_POPULATE | MAP_FIXED, fd, 0); + if (addr == MAP_FAILED) { + RTE_LOG(ERR, EAL, "Couldn't remap '%s': %s\n", + hfile->filepath, strerror(errno)); + close(fd); + return -1; + } + + /* we have a new address, so unmap previous one */ +#ifndef RTE_ARCH_64 + /* in 32-bit legacy mode, we have already unmapped the page */ + if (!internal_config.legacy_mem) + munmap(hfile->orig_va, page_sz); +#else + munmap(hfile->orig_va, page_sz); +#endif + + hfile->orig_va = NULL; + hfile->final_va = addr; + + /* rewrite physical addresses in IOVA as VA mode */ + if (rte_eal_iova_mode() == RTE_IOVA_VA) + hfile->physaddr = (uintptr_t)addr; + + /* set up memseg data */ + ms->addr = addr; + ms->hugepage_sz = page_sz; + ms->len = memseg_len; + ms->iova = hfile->physaddr; + ms->socket_id = hfile->socket_id; + ms->nchannel = rte_memory_get_nchannel(); + ms->nrank = rte_memory_get_nrank(); + + rte_fbarray_set_used(arr, ms_idx); + + close(fd); + } + RTE_LOG(DEBUG, EAL, "Allocated %" PRIu64 "M on socket %i\n", + (seg_len * page_sz) >> 20, socket_id); + return 0; +} + +static uint64_t +get_mem_amount(uint64_t page_sz, uint64_t max_mem) +{ + uint64_t area_sz, max_pages; + + /* limit to RTE_MAX_MEMSEG_PER_LIST pages or RTE_MAX_MEM_MB_PER_LIST */ + max_pages = RTE_MAX_MEMSEG_PER_LIST; + max_mem = RTE_MIN((uint64_t)RTE_MAX_MEM_MB_PER_LIST << 20, max_mem); + + area_sz = RTE_MIN(page_sz * max_pages, max_mem); + + /* make sure the list isn't smaller than the page size */ + area_sz = RTE_MAX(area_sz, page_sz); + + return RTE_ALIGN(area_sz, page_sz); +} + +static int +free_memseg_list(struct rte_memseg_list *msl) +{ + if (rte_fbarray_destroy(&msl->memseg_arr)) { + RTE_LOG(ERR, EAL, "Cannot destroy memseg list\n"); + return -1; + } + memset(msl, 0, sizeof(*msl)); + return 0; +} + +#define MEMSEG_LIST_FMT "memseg-%" PRIu64 "k-%i-%i" +static int +alloc_memseg_list(struct rte_memseg_list *msl, uint64_t page_sz, + int n_segs, int socket_id, int type_msl_idx) +{ + char name[RTE_FBARRAY_NAME_LEN]; + + snprintf(name, sizeof(name), MEMSEG_LIST_FMT, page_sz >> 10, socket_id, + type_msl_idx); + if (rte_fbarray_init(&msl->memseg_arr, name, n_segs, + sizeof(struct rte_memseg))) { + RTE_LOG(ERR, EAL, "Cannot allocate memseg list: %s\n", + rte_strerror(rte_errno)); + return -1; + } + + msl->page_sz = page_sz; + msl->socket_id = socket_id; + msl->base_va = NULL; + + RTE_LOG(DEBUG, EAL, "Memseg list allocated: 0x%zxkB at socket %i\n", + (size_t)page_sz >> 10, socket_id); + + return 0; +} + +static int +alloc_va_space(struct rte_memseg_list *msl) +{ + uint64_t page_sz; + size_t mem_sz; + void *addr; + int flags = 0; + +#ifdef RTE_ARCH_PPC_64 + flags |= MAP_HUGETLB; +#endif + + page_sz = msl->page_sz; + mem_sz = page_sz * msl->memseg_arr.len; + + addr = eal_get_virtual_area(msl->base_va, &mem_sz, page_sz, 0, flags); + if (addr == NULL) { + if (rte_errno == EADDRNOTAVAIL) + RTE_LOG(ERR, EAL, "Could not mmap %llu bytes at [%p] - please use '--base-virtaddr' option\n", + (unsigned long long)mem_sz, msl->base_va); + else + RTE_LOG(ERR, EAL, "Cannot reserve memory\n"); + return -1; + } + msl->base_va = addr; + + return 0; +} + +/* + * Our VA space is not preallocated yet, so preallocate it here. We need to know + * how many segments there are in order to map all pages into one address space, + * and leave appropriate holes between segments so that rte_malloc does not + * concatenate them into one big segment. + * + * we also need to unmap original pages to free up address space. + */ +static int __rte_unused +prealloc_segments(struct hugepage_file *hugepages, int n_pages) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + int cur_page, seg_start_page, end_seg, new_memseg; + unsigned int hpi_idx, socket, i; + int n_contig_segs, n_segs; + int msl_idx; + + /* before we preallocate segments, we need to free up our VA space. + * we're not removing files, and we already have information about + * PA-contiguousness, so it is safe to unmap everything. + */ + for (cur_page = 0; cur_page < n_pages; cur_page++) { + struct hugepage_file *hpi = &hugepages[cur_page]; + munmap(hpi->orig_va, hpi->size); + hpi->orig_va = NULL; + } + + /* we cannot know how many page sizes and sockets we have discovered, so + * loop over all of them + */ + for (hpi_idx = 0; hpi_idx < internal_config.num_hugepage_sizes; + hpi_idx++) { + uint64_t page_sz = + internal_config.hugepage_info[hpi_idx].hugepage_sz; + + for (i = 0; i < rte_socket_count(); i++) { + struct rte_memseg_list *msl; + + socket = rte_socket_id_by_idx(i); + n_contig_segs = 0; + n_segs = 0; + seg_start_page = -1; + + for (cur_page = 0; cur_page < n_pages; cur_page++) { + struct hugepage_file *prev, *cur; + int prev_seg_start_page = -1; + + cur = &hugepages[cur_page]; + prev = cur_page == 0 ? NULL : + &hugepages[cur_page - 1]; + + new_memseg = 0; + end_seg = 0; + + if (cur->size == 0) + end_seg = 1; + else if (cur->socket_id != (int) socket) + end_seg = 1; + else if (cur->size != page_sz) + end_seg = 1; + else if (cur_page == 0) + new_memseg = 1; +#ifdef RTE_ARCH_PPC_64 + /* On PPC64 architecture, the mmap always start + * from higher address to lower address. Here, + * physical addresses are in descending order. + */ + else if ((prev->physaddr - cur->physaddr) != + cur->size) + new_memseg = 1; +#else + else if ((cur->physaddr - prev->physaddr) != + cur->size) + new_memseg = 1; +#endif + if (new_memseg) { + /* if we're already inside a segment, + * new segment means end of current one + */ + if (seg_start_page != -1) { + end_seg = 1; + prev_seg_start_page = + seg_start_page; + } + seg_start_page = cur_page; + } + + if (end_seg) { + if (prev_seg_start_page != -1) { + /* we've found a new segment */ + n_contig_segs++; + n_segs += cur_page - + prev_seg_start_page; + } else if (seg_start_page != -1) { + /* we didn't find new segment, + * but did end current one + */ + n_contig_segs++; + n_segs += cur_page - + seg_start_page; + seg_start_page = -1; + continue; + } else { + /* we're skipping this page */ + continue; + } + } + /* segment continues */ + } + /* check if we missed last segment */ + if (seg_start_page != -1) { + n_contig_segs++; + n_segs += cur_page - seg_start_page; + } + + /* if no segments were found, do not preallocate */ + if (n_segs == 0) + continue; + + /* we now have total number of pages that we will + * allocate for this segment list. add separator pages + * to the total count, and preallocate VA space. + */ + n_segs += n_contig_segs - 1; + + /* now, preallocate VA space for these segments */ + + /* first, find suitable memseg list for this */ + for (msl_idx = 0; msl_idx < RTE_MAX_MEMSEG_LISTS; + msl_idx++) { + msl = &mcfg->memsegs[msl_idx]; + + if (msl->base_va != NULL) + continue; + break; + } + if (msl_idx == RTE_MAX_MEMSEG_LISTS) { + RTE_LOG(ERR, EAL, "Not enough space in memseg lists, please increase %s\n", + RTE_STR(CONFIG_RTE_MAX_MEMSEG_LISTS)); + return -1; + } + + /* now, allocate fbarray itself */ + if (alloc_memseg_list(msl, page_sz, n_segs, socket, + msl_idx) < 0) + return -1; + + /* finally, allocate VA space */ + if (alloc_va_space(msl) < 0) + return -1; + } + } + return 0; +} + +/* + * We cannot reallocate memseg lists on the fly because PPC64 stores pages + * backwards, therefore we have to process the entire memseg first before + * remapping it into memseg list VA space. + */ +static int +remap_needed_hugepages(struct hugepage_file *hugepages, int n_pages) +{ + int cur_page, seg_start_page, new_memseg, ret; + + seg_start_page = 0; + for (cur_page = 0; cur_page < n_pages; cur_page++) { + struct hugepage_file *prev, *cur; + + new_memseg = 0; + + cur = &hugepages[cur_page]; + prev = cur_page == 0 ? NULL : &hugepages[cur_page - 1]; + + /* if size is zero, no more pages left */ + if (cur->size == 0) + break; + + if (cur_page == 0) + new_memseg = 1; + else if (cur->socket_id != prev->socket_id) + new_memseg = 1; + else if (cur->size != prev->size) + new_memseg = 1; +#ifdef RTE_ARCH_PPC_64 + /* On PPC64 architecture, the mmap always start from higher + * address to lower address. Here, physical addresses are in + * descending order. + */ + else if ((prev->physaddr - cur->physaddr) != cur->size) + new_memseg = 1; +#else + else if ((cur->physaddr - prev->physaddr) != cur->size) + new_memseg = 1; +#endif + + if (new_memseg) { + /* if this isn't the first time, remap segment */ + if (cur_page != 0) { + ret = remap_segment(hugepages, seg_start_page, + cur_page); + if (ret != 0) + return -1; + } + /* remember where we started */ + seg_start_page = cur_page; + } + /* continuation of previous memseg */ + } + /* we were stopped, but we didn't remap the last segment, do it now */ + if (cur_page != 0) { + ret = remap_segment(hugepages, seg_start_page, + cur_page); + if (ret != 0) + return -1; + } + return 0; +} + static inline uint64_t get_socket_mem_size(int socket) { @@ -788,8 +1088,7 @@ get_socket_mem_size(int socket) for (i = 0; i < internal_config.num_hugepage_sizes; i++){ struct hugepage_info *hpi = &internal_config.hugepage_info[i]; - if (hpi->hugedir != NULL) - size += hpi->hugepage_sz * hpi->num_pages[socket]; + size += hpi->hugepage_sz * hpi->num_pages[socket]; } return size; @@ -818,8 +1117,10 @@ calc_num_pages_per_socket(uint64_t * memory, /* if specific memory amounts per socket weren't requested */ if (internal_config.force_sockets == 0) { + size_t total_size; +#ifdef RTE_ARCH_64 int cpu_per_socket[RTE_MAX_NUMA_NODES]; - size_t default_size, total_size; + size_t default_size; unsigned lcore_id; /* Compute number of cores per socket */ @@ -837,7 +1138,7 @@ calc_num_pages_per_socket(uint64_t * memory, /* Set memory amount per socket */ default_size = (internal_config.memory * cpu_per_socket[socket]) - / rte_lcore_count(); + / rte_lcore_count(); /* Limit to maximum available memory on socket */ default_size = RTE_MIN(default_size, get_socket_mem_size(socket)); @@ -854,18 +1155,40 @@ calc_num_pages_per_socket(uint64_t * memory, for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_size != 0; socket++) { /* take whatever is available */ default_size = RTE_MIN(get_socket_mem_size(socket) - memory[socket], - total_size); + total_size); /* Update sizes */ memory[socket] += default_size; total_size -= default_size; } +#else + /* in 32-bit mode, allocate all of the memory only on master + * lcore socket + */ + total_size = internal_config.memory; + for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_size != 0; + socket++) { + struct rte_config *cfg = rte_eal_get_configuration(); + unsigned int master_lcore_socket; + + master_lcore_socket = + rte_lcore_to_socket_id(cfg->master_lcore); + + if (master_lcore_socket != socket) + continue; + + /* Update sizes */ + memory[socket] = total_size; + break; + } +#endif } for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_mem != 0; socket++) { /* skips if the memory on specific socket wasn't requested */ for (i = 0; i < num_hp_info && memory[socket] != 0; i++){ - hp_used[i].hugedir = hp_info[i].hugedir; + strlcpy(hp_used[i].hugedir, hp_info[i].hugedir, + sizeof(hp_used[i].hugedir)); hp_used[i].num_pages[socket] = RTE_MIN( memory[socket] / hp_info[i].hugepage_sz, hp_info[i].num_pages[socket]); @@ -907,7 +1230,8 @@ calc_num_pages_per_socket(uint64_t * memory, } } /* if we didn't satisfy all memory requirements per socket */ - if (memory[socket] > 0) { + if (memory[socket] > 0 && + internal_config.socket_mem[socket] != 0) { /* to prevent icc errors */ requested = (unsigned) (internal_config.socket_mem[socket] / 0x100000); @@ -939,7 +1263,7 @@ eal_get_hugepage_mem_size(void) for (i = 0; i < internal_config.num_hugepage_sizes; i++) { struct hugepage_info *hpi = &internal_config.hugepage_info[i]; - if (hpi->hugedir != NULL) { + if (strnlen(hpi->hugedir, sizeof(hpi->hugedir)) != 0) { for (j = 0; j < RTE_MAX_NUMA_NODES; j++) { size += hpi->hugepage_sz * hpi->num_pages[j]; } @@ -987,17 +1311,19 @@ huge_recover_sigbus(void) * 6. unmap the first mapping * 7. fill memsegs in configuration with contiguous zones */ -int -rte_eal_hugepage_init(void) +static int +eal_legacy_hugepage_init(void) { struct rte_mem_config *mcfg; struct hugepage_file *hugepage = NULL, *tmp_hp = NULL; struct hugepage_info used_hp[MAX_HUGEPAGE_SIZES]; + struct rte_fbarray *arr; + struct rte_memseg *ms; uint64_t memory[RTE_MAX_NUMA_NODES]; unsigned hp_offset; - int i, j, new_memseg; + int i, j; int nr_hugefiles, nr_hugepages = 0; void *addr; @@ -1010,21 +1336,54 @@ rte_eal_hugepage_init(void) /* hugetlbfs can be disabled */ if (internal_config.no_hugetlbfs) { + struct rte_memseg_list *msl; + uint64_t page_sz; + int n_segs, cur_seg; + + /* nohuge mode is legacy mode */ + internal_config.legacy_mem = 1; + + /* create a memseg list */ + msl = &mcfg->memsegs[0]; + + page_sz = RTE_PGSIZE_4K; + n_segs = internal_config.memory / page_sz; + + if (rte_fbarray_init(&msl->memseg_arr, "nohugemem", n_segs, + sizeof(struct rte_memseg))) { + RTE_LOG(ERR, EAL, "Cannot allocate memseg list\n"); + return -1; + } + addr = mmap(NULL, internal_config.memory, PROT_READ | PROT_WRITE, - MAP_PRIVATE | MAP_ANONYMOUS, 0, 0); + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); if (addr == MAP_FAILED) { RTE_LOG(ERR, EAL, "%s: mmap() failed: %s\n", __func__, strerror(errno)); return -1; } - if (rte_eal_iova_mode() == RTE_IOVA_VA) - mcfg->memseg[0].iova = (uintptr_t)addr; - else - mcfg->memseg[0].iova = RTE_BAD_IOVA; - mcfg->memseg[0].addr = addr; - mcfg->memseg[0].hugepage_sz = RTE_PGSIZE_4K; - mcfg->memseg[0].len = internal_config.memory; - mcfg->memseg[0].socket_id = 0; + msl->base_va = addr; + msl->page_sz = page_sz; + msl->socket_id = 0; + + /* populate memsegs. each memseg is one page long */ + for (cur_seg = 0; cur_seg < n_segs; cur_seg++) { + arr = &msl->memseg_arr; + + ms = rte_fbarray_get(arr, cur_seg); + if (rte_eal_iova_mode() == RTE_IOVA_VA) + ms->iova = (uintptr_t)addr; + else + ms->iova = RTE_BAD_IOVA; + ms->addr = addr; + ms->hugepage_sz = page_sz; + ms->socket_id = 0; + ms->len = page_sz; + + rte_fbarray_set_used(arr, cur_seg); + + addr = RTE_PTR_ADD(addr, (size_t)page_sz); + } return 0; } @@ -1057,7 +1416,6 @@ rte_eal_hugepage_init(void) for (i = 0; i < RTE_MAX_NUMA_NODES; i++) memory[i] = internal_config.socket_mem[i]; - /* map all hugepages and sort them */ for (i = 0; i < (int)internal_config.num_hugepage_sizes; i ++){ unsigned pages_old, pages_new; @@ -1075,8 +1433,7 @@ rte_eal_hugepage_init(void) /* map all hugepages available */ pages_old = hpi->num_pages[0]; - pages_new = map_all_hugepages(&tmp_hp[hp_offset], hpi, - memory, 1); + pages_new = map_all_hugepages(&tmp_hp[hp_offset], hpi, memory); if (pages_new < pages_old) { RTE_LOG(DEBUG, EAL, "%d not %d hugepages of size %u MB allocated\n", @@ -1091,7 +1448,8 @@ rte_eal_hugepage_init(void) continue; } - if (phys_addrs_available) { + if (phys_addrs_available && + rte_eal_iova_mode() != RTE_IOVA_VA) { /* find physical addresses for each hugepage */ if (find_physaddrs(&tmp_hp[hp_offset], hpi) < 0) { RTE_LOG(DEBUG, EAL, "Failed to find phys addr " @@ -1118,18 +1476,6 @@ rte_eal_hugepage_init(void) qsort(&tmp_hp[hp_offset], hpi->num_pages[0], sizeof(struct hugepage_file), cmp_physaddr); - /* remap all hugepages */ - if (map_all_hugepages(&tmp_hp[hp_offset], hpi, NULL, 0) != - hpi->num_pages[0]) { - RTE_LOG(ERR, EAL, "Failed to remap %u MB pages\n", - (unsigned)(hpi->hugepage_sz / 0x100000)); - goto fail; - } - - /* unmap original mappings */ - if (unmap_all_hugepages_orig(&tmp_hp[hp_offset], hpi) < 0) - goto fail; - /* we have processed a num of hugepages of this size, so inc offset */ hp_offset += hpi->num_pages[0]; } @@ -1191,7 +1537,7 @@ rte_eal_hugepage_init(void) } /* create shared memory */ - hugepage = create_shared_memory(eal_hugepage_info_path(), + hugepage = create_shared_memory(eal_hugepage_data_path(), nr_hugefiles * sizeof(struct hugepage_file)); if (hugepage == NULL) { @@ -1212,7 +1558,7 @@ rte_eal_hugepage_init(void) /* * copy stuff from malloc'd hugepage* to the actual shared memory. - * this procedure only copies those hugepages that have final_va + * this procedure only copies those hugepages that have orig_va * not NULL. has overflow protection. */ if (copy_hugepages_to_shared_mem(hugepage, nr_hugefiles, @@ -1221,6 +1567,23 @@ rte_eal_hugepage_init(void) goto fail; } +#ifndef RTE_ARCH_64 + /* for legacy 32-bit mode, we did not preallocate VA space, so do it */ + if (internal_config.legacy_mem && + prealloc_segments(hugepage, nr_hugefiles)) { + RTE_LOG(ERR, EAL, "Could not preallocate VA space for hugepages\n"); + goto fail; + } +#endif + + /* remap all pages we do need into memseg list VA space, so that those + * pages become first-class citizens in DPDK memory subsystem + */ + if (remap_needed_hugepages(hugepage, nr_hugefiles)) { + RTE_LOG(ERR, EAL, "Couldn't remap hugepage files into memseg lists\n"); + goto fail; + } + /* free the hugepage backing files */ if (internal_config.hugepage_unlink && unlink_hugepage_files(tmp_hp, internal_config.num_hugepage_sizes) < 0) { @@ -1232,75 +1595,30 @@ rte_eal_hugepage_init(void) free(tmp_hp); tmp_hp = NULL; - /* first memseg index shall be 0 after incrementing it below */ - j = -1; - for (i = 0; i < nr_hugefiles; i++) { - new_memseg = 0; - - /* if this is a new section, create a new memseg */ - if (i == 0) - new_memseg = 1; - else if (hugepage[i].socket_id != hugepage[i-1].socket_id) - new_memseg = 1; - else if (hugepage[i].size != hugepage[i-1].size) - new_memseg = 1; - -#ifdef RTE_ARCH_PPC_64 - /* On PPC64 architecture, the mmap always start from higher - * virtual address to lower address. Here, both the physical - * address and virtual address are in descending order */ - else if ((hugepage[i-1].physaddr - hugepage[i].physaddr) != - hugepage[i].size) - new_memseg = 1; - else if (((unsigned long)hugepage[i-1].final_va - - (unsigned long)hugepage[i].final_va) != hugepage[i].size) - new_memseg = 1; -#else - else if ((hugepage[i].physaddr - hugepage[i-1].physaddr) != - hugepage[i].size) - new_memseg = 1; - else if (((unsigned long)hugepage[i].final_va - - (unsigned long)hugepage[i-1].final_va) != hugepage[i].size) - new_memseg = 1; -#endif + munmap(hugepage, nr_hugefiles * sizeof(struct hugepage_file)); - if (new_memseg) { - j += 1; - if (j == RTE_MAX_MEMSEG) - break; + /* we're not going to allocate more pages, so release VA space for + * unused memseg lists + */ + for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) { + struct rte_memseg_list *msl = &mcfg->memsegs[i]; + size_t mem_sz; - mcfg->memseg[j].iova = hugepage[i].physaddr; - mcfg->memseg[j].addr = hugepage[i].final_va; - mcfg->memseg[j].len = hugepage[i].size; - mcfg->memseg[j].socket_id = hugepage[i].socket_id; - mcfg->memseg[j].hugepage_sz = hugepage[i].size; - } - /* continuation of previous memseg */ - else { -#ifdef RTE_ARCH_PPC_64 - /* Use the phy and virt address of the last page as segment - * address for IBM Power architecture */ - mcfg->memseg[j].iova = hugepage[i].physaddr; - mcfg->memseg[j].addr = hugepage[i].final_va; -#endif - mcfg->memseg[j].len += mcfg->memseg[j].hugepage_sz; - } - hugepage[i].memseg_id = j; - } + /* skip inactive lists */ + if (msl->base_va == NULL) + continue; + /* skip lists where there is at least one page allocated */ + if (msl->memseg_arr.count > 0) + continue; + /* this is an unused list, deallocate it */ + mem_sz = (size_t)msl->page_sz * msl->memseg_arr.len; + munmap(msl->base_va, mem_sz); + msl->base_va = NULL; - if (i < nr_hugefiles) { - RTE_LOG(ERR, EAL, "Can only reserve %d pages " - "from %d requested\n" - "Current %s=%d is not enough\n" - "Please either increase it or request less amount " - "of memory.\n", - i, nr_hugefiles, RTE_STR(CONFIG_RTE_MAX_MEMSEG), - RTE_MAX_MEMSEG); - goto fail; + /* destroy backing fbarray */ + rte_fbarray_destroy(&msl->memseg_arr); } - munmap(hugepage, nr_hugefiles * sizeof(struct hugepage_file)); - return 0; fail: @@ -1312,6 +1630,125 @@ fail: return -1; } +static int __rte_unused +hugepage_count_walk(const struct rte_memseg_list *msl, void *arg) +{ + struct hugepage_info *hpi = arg; + + if (msl->page_sz != hpi->hugepage_sz) + return 0; + + hpi->num_pages[msl->socket_id] += msl->memseg_arr.len; + return 0; +} + +static int +limits_callback(int socket_id, size_t cur_limit, size_t new_len) +{ + RTE_SET_USED(socket_id); + RTE_SET_USED(cur_limit); + RTE_SET_USED(new_len); + return -1; +} + +static int +eal_hugepage_init(void) +{ + struct hugepage_info used_hp[MAX_HUGEPAGE_SIZES]; + uint64_t memory[RTE_MAX_NUMA_NODES]; + int hp_sz_idx, socket_id; + + test_phys_addrs_available(); + + memset(used_hp, 0, sizeof(used_hp)); + + for (hp_sz_idx = 0; + hp_sz_idx < (int) internal_config.num_hugepage_sizes; + hp_sz_idx++) { +#ifndef RTE_ARCH_64 + struct hugepage_info dummy; + unsigned int i; +#endif + /* also initialize used_hp hugepage sizes in used_hp */ + struct hugepage_info *hpi; + hpi = &internal_config.hugepage_info[hp_sz_idx]; + used_hp[hp_sz_idx].hugepage_sz = hpi->hugepage_sz; + +#ifndef RTE_ARCH_64 + /* for 32-bit, limit number of pages on socket to whatever we've + * preallocated, as we cannot allocate more. + */ + memset(&dummy, 0, sizeof(dummy)); + dummy.hugepage_sz = hpi->hugepage_sz; + if (rte_memseg_list_walk(hugepage_count_walk, &dummy) < 0) + return -1; + + for (i = 0; i < RTE_DIM(dummy.num_pages); i++) { + hpi->num_pages[i] = RTE_MIN(hpi->num_pages[i], + dummy.num_pages[i]); + } +#endif + } + + /* make a copy of socket_mem, needed for balanced allocation. */ + for (hp_sz_idx = 0; hp_sz_idx < RTE_MAX_NUMA_NODES; hp_sz_idx++) + memory[hp_sz_idx] = internal_config.socket_mem[hp_sz_idx]; + + /* calculate final number of pages */ + if (calc_num_pages_per_socket(memory, + internal_config.hugepage_info, used_hp, + internal_config.num_hugepage_sizes) < 0) + return -1; + + for (hp_sz_idx = 0; + hp_sz_idx < (int)internal_config.num_hugepage_sizes; + hp_sz_idx++) { + for (socket_id = 0; socket_id < RTE_MAX_NUMA_NODES; + socket_id++) { + struct rte_memseg **pages; + struct hugepage_info *hpi = &used_hp[hp_sz_idx]; + unsigned int num_pages = hpi->num_pages[socket_id]; + int num_pages_alloc, i; + + if (num_pages == 0) + continue; + + pages = malloc(sizeof(*pages) * num_pages); + + RTE_LOG(DEBUG, EAL, "Allocating %u pages of size %" PRIu64 "M on socket %i\n", + num_pages, hpi->hugepage_sz >> 20, socket_id); + + num_pages_alloc = eal_memalloc_alloc_seg_bulk(pages, + num_pages, hpi->hugepage_sz, + socket_id, true); + if (num_pages_alloc < 0) { + free(pages); + return -1; + } + + /* mark preallocated pages as unfreeable */ + for (i = 0; i < num_pages_alloc; i++) { + struct rte_memseg *ms = pages[i]; + ms->flags |= RTE_MEMSEG_FLAG_DO_NOT_FREE; + } + free(pages); + } + } + /* if socket limits were specified, set them */ + if (internal_config.force_socket_limits) { + unsigned int i; + for (i = 0; i < RTE_MAX_NUMA_NODES; i++) { + uint64_t limit = internal_config.socket_limit[i]; + if (limit == 0) + continue; + if (rte_mem_alloc_validator_register("socket-limit", + limits_callback, i, limit)) + RTE_LOG(ERR, EAL, "Failed to register socket limits validator callback\n"); + } + } + return 0; +} + /* * uses fstat to report the size of a file on disk */ @@ -1330,16 +1767,15 @@ getFileSize(int fd) * configuration and finds the hugepages which form that segment, mapping them * in order to form a contiguous block in the virtual memory space */ -int -rte_eal_hugepage_attach(void) +static int +eal_legacy_hugepage_attach(void) { - const struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; struct hugepage_file *hp = NULL; - unsigned num_hp = 0; - unsigned i, s = 0; /* s used to track the segment number */ - unsigned max_seg = RTE_MAX_MEMSEG; + unsigned int num_hp = 0; + unsigned int i = 0; + unsigned int cur_seg; off_t size = 0; - int fd, fd_zero = -1, fd_hugepage = -1; + int fd, fd_hugepage = -1; if (aslr_enabled() > 0) { RTE_LOG(WARNING, EAL, "WARNING: Address Space Layout Randomization " @@ -1350,139 +1786,429 @@ rte_eal_hugepage_attach(void) test_phys_addrs_available(); - fd_zero = open("/dev/zero", O_RDONLY); - if (fd_zero < 0) { - RTE_LOG(ERR, EAL, "Could not open /dev/zero\n"); - goto error; - } - fd_hugepage = open(eal_hugepage_info_path(), O_RDONLY); + fd_hugepage = open(eal_hugepage_data_path(), O_RDONLY); if (fd_hugepage < 0) { - RTE_LOG(ERR, EAL, "Could not open %s\n", eal_hugepage_info_path()); + RTE_LOG(ERR, EAL, "Could not open %s\n", + eal_hugepage_data_path()); goto error; } - /* map all segments into memory to make sure we get the addrs */ - for (s = 0; s < RTE_MAX_MEMSEG; ++s) { - void *base_addr; - - /* - * the first memory segment with len==0 is the one that - * follows the last valid segment. - */ - if (mcfg->memseg[s].len == 0) - break; - - /* - * fdzero is mmapped to get a contiguous block of virtual - * addresses of the appropriate memseg size. - * use mmap to get identical addresses as the primary process. - */ - base_addr = mmap(mcfg->memseg[s].addr, mcfg->memseg[s].len, - PROT_READ, -#ifdef RTE_ARCH_PPC_64 - MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB, -#else - MAP_PRIVATE, -#endif - fd_zero, 0); - if (base_addr == MAP_FAILED || - base_addr != mcfg->memseg[s].addr) { - max_seg = s; - if (base_addr != MAP_FAILED) { - /* errno is stale, don't use */ - RTE_LOG(ERR, EAL, "Could not mmap %llu bytes " - "in /dev/zero at [%p], got [%p] - " - "please use '--base-virtaddr' option\n", - (unsigned long long)mcfg->memseg[s].len, - mcfg->memseg[s].addr, base_addr); - munmap(base_addr, mcfg->memseg[s].len); - } else { - RTE_LOG(ERR, EAL, "Could not mmap %llu bytes " - "in /dev/zero at [%p]: '%s'\n", - (unsigned long long)mcfg->memseg[s].len, - mcfg->memseg[s].addr, strerror(errno)); - } - if (aslr_enabled() > 0) { - RTE_LOG(ERR, EAL, "It is recommended to " - "disable ASLR in the kernel " - "and retry running both primary " - "and secondary processes\n"); - } - goto error; - } - } - size = getFileSize(fd_hugepage); hp = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd_hugepage, 0); if (hp == MAP_FAILED) { - RTE_LOG(ERR, EAL, "Could not mmap %s\n", eal_hugepage_info_path()); + RTE_LOG(ERR, EAL, "Could not mmap %s\n", + eal_hugepage_data_path()); goto error; } num_hp = size / sizeof(struct hugepage_file); RTE_LOG(DEBUG, EAL, "Analysing %u files\n", num_hp); - s = 0; - while (s < RTE_MAX_MEMSEG && mcfg->memseg[s].len > 0){ - void *addr, *base_addr; - uintptr_t offset = 0; - size_t mapping_size; - /* - * free previously mapped memory so we can map the - * hugepages into the space - */ - base_addr = mcfg->memseg[s].addr; - munmap(base_addr, mcfg->memseg[s].len); - - /* find the hugepages for this segment and map them - * we don't need to worry about order, as the server sorted the - * entries before it did the second mmap of them */ - for (i = 0; i < num_hp && offset < mcfg->memseg[s].len; i++){ - if (hp[i].memseg_id == (int)s){ - fd = open(hp[i].filepath, O_RDWR); - if (fd < 0) { - RTE_LOG(ERR, EAL, "Could not open %s\n", - hp[i].filepath); - goto error; - } - mapping_size = hp[i].size; - addr = mmap(RTE_PTR_ADD(base_addr, offset), - mapping_size, PROT_READ | PROT_WRITE, - MAP_SHARED, fd, 0); - close(fd); /* close file both on success and on failure */ - if (addr == MAP_FAILED || - addr != RTE_PTR_ADD(base_addr, offset)) { - RTE_LOG(ERR, EAL, "Could not mmap %s\n", - hp[i].filepath); - goto error; - } - offset+=mapping_size; - } + /* map all segments into memory to make sure we get the addrs. the + * segments themselves are already in memseg list (which is shared and + * has its VA space already preallocated), so we just need to map + * everything into correct addresses. + */ + for (i = 0; i < num_hp; i++) { + struct hugepage_file *hf = &hp[i]; + size_t map_sz = hf->size; + void *map_addr = hf->final_va; + + /* if size is zero, no more pages left */ + if (map_sz == 0) + break; + + fd = open(hf->filepath, O_RDWR); + if (fd < 0) { + RTE_LOG(ERR, EAL, "Could not open %s: %s\n", + hf->filepath, strerror(errno)); + goto error; + } + + map_addr = mmap(map_addr, map_sz, PROT_READ | PROT_WRITE, + MAP_SHARED | MAP_FIXED, fd, 0); + if (map_addr == MAP_FAILED) { + RTE_LOG(ERR, EAL, "Could not map %s: %s\n", + hf->filepath, strerror(errno)); + close(fd); + goto error; + } + + /* set shared lock on the file. */ + if (flock(fd, LOCK_SH) < 0) { + RTE_LOG(DEBUG, EAL, "%s(): Locking file failed: %s\n", + __func__, strerror(errno)); + close(fd); + goto error; } - RTE_LOG(DEBUG, EAL, "Mapped segment %u of size 0x%llx\n", s, - (unsigned long long)mcfg->memseg[s].len); - s++; + + close(fd); } /* unmap the hugepage config file, since we are done using it */ munmap(hp, size); - close(fd_zero); close(fd_hugepage); return 0; error: - for (i = 0; i < max_seg && mcfg->memseg[i].len > 0; i++) - munmap(mcfg->memseg[i].addr, mcfg->memseg[i].len); + /* map all segments into memory to make sure we get the addrs */ + cur_seg = 0; + for (cur_seg = 0; cur_seg < i; cur_seg++) { + struct hugepage_file *hf = &hp[i]; + size_t map_sz = hf->size; + void *map_addr = hf->final_va; + + munmap(map_addr, map_sz); + } if (hp != NULL && hp != MAP_FAILED) munmap(hp, size); - if (fd_zero >= 0) - close(fd_zero); if (fd_hugepage >= 0) close(fd_hugepage); return -1; } +static int +eal_hugepage_attach(void) +{ + if (eal_memalloc_sync_with_primary()) { + RTE_LOG(ERR, EAL, "Could not map memory from primary process\n"); + if (aslr_enabled() > 0) + RTE_LOG(ERR, EAL, "It is recommended to disable ASLR in the kernel and retry running both primary and secondary processes\n"); + return -1; + } + return 0; +} + +int +rte_eal_hugepage_init(void) +{ + return internal_config.legacy_mem ? + eal_legacy_hugepage_init() : + eal_hugepage_init(); +} + +int +rte_eal_hugepage_attach(void) +{ + return internal_config.legacy_mem ? + eal_legacy_hugepage_attach() : + eal_hugepage_attach(); +} + int rte_eal_using_phys_addrs(void) { return phys_addrs_available; } + +static int __rte_unused +memseg_primary_init_32(void) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + int active_sockets, hpi_idx, msl_idx = 0; + unsigned int socket_id, i; + struct rte_memseg_list *msl; + uint64_t extra_mem_per_socket, total_extra_mem, total_requested_mem; + uint64_t max_mem; + + /* no-huge does not need this at all */ + if (internal_config.no_hugetlbfs) + return 0; + + /* this is a giant hack, but desperate times call for desperate + * measures. in legacy 32-bit mode, we cannot preallocate VA space, + * because having upwards of 2 gigabytes of VA space already mapped will + * interfere with our ability to map and sort hugepages. + * + * therefore, in legacy 32-bit mode, we will be initializing memseg + * lists much later - in eal_memory.c, right after we unmap all the + * unneeded pages. this will not affect secondary processes, as those + * should be able to mmap the space without (too many) problems. + */ + if (internal_config.legacy_mem) + return 0; + + /* 32-bit mode is a very special case. we cannot know in advance where + * the user will want to allocate their memory, so we have to do some + * heuristics. + */ + active_sockets = 0; + total_requested_mem = 0; + if (internal_config.force_sockets) + for (i = 0; i < rte_socket_count(); i++) { + uint64_t mem; + + socket_id = rte_socket_id_by_idx(i); + mem = internal_config.socket_mem[socket_id]; + + if (mem == 0) + continue; + + active_sockets++; + total_requested_mem += mem; + } + else + total_requested_mem = internal_config.memory; + + max_mem = (uint64_t)RTE_MAX_MEM_MB << 20; + if (total_requested_mem > max_mem) { + RTE_LOG(ERR, EAL, "Invalid parameters: 32-bit process can at most use %uM of memory\n", + (unsigned int)(max_mem >> 20)); + return -1; + } + total_extra_mem = max_mem - total_requested_mem; + extra_mem_per_socket = active_sockets == 0 ? total_extra_mem : + total_extra_mem / active_sockets; + + /* the allocation logic is a little bit convoluted, but here's how it + * works, in a nutshell: + * - if user hasn't specified on which sockets to allocate memory via + * --socket-mem, we allocate all of our memory on master core socket. + * - if user has specified sockets to allocate memory on, there may be + * some "unused" memory left (e.g. if user has specified --socket-mem + * such that not all memory adds up to 2 gigabytes), so add it to all + * sockets that are in use equally. + * + * page sizes are sorted by size in descending order, so we can safely + * assume that we dispense with bigger page sizes first. + */ + + /* create memseg lists */ + for (i = 0; i < rte_socket_count(); i++) { + int hp_sizes = (int) internal_config.num_hugepage_sizes; + uint64_t max_socket_mem, cur_socket_mem; + unsigned int master_lcore_socket; + struct rte_config *cfg = rte_eal_get_configuration(); + bool skip; + + socket_id = rte_socket_id_by_idx(i); + +#ifndef RTE_EAL_NUMA_AWARE_HUGEPAGES + if (socket_id > 0) + break; +#endif + + /* if we didn't specifically request memory on this socket */ + skip = active_sockets != 0 && + internal_config.socket_mem[socket_id] == 0; + /* ...or if we didn't specifically request memory on *any* + * socket, and this is not master lcore + */ + master_lcore_socket = rte_lcore_to_socket_id(cfg->master_lcore); + skip |= active_sockets == 0 && socket_id != master_lcore_socket; + + if (skip) { + RTE_LOG(DEBUG, EAL, "Will not preallocate memory on socket %u\n", + socket_id); + continue; + } + + /* max amount of memory on this socket */ + max_socket_mem = (active_sockets != 0 ? + internal_config.socket_mem[socket_id] : + internal_config.memory) + + extra_mem_per_socket; + cur_socket_mem = 0; + + for (hpi_idx = 0; hpi_idx < hp_sizes; hpi_idx++) { + uint64_t max_pagesz_mem, cur_pagesz_mem = 0; + uint64_t hugepage_sz; + struct hugepage_info *hpi; + int type_msl_idx, max_segs, total_segs = 0; + + hpi = &internal_config.hugepage_info[hpi_idx]; + hugepage_sz = hpi->hugepage_sz; + + /* check if pages are actually available */ + if (hpi->num_pages[socket_id] == 0) + continue; + + max_segs = RTE_MAX_MEMSEG_PER_TYPE; + max_pagesz_mem = max_socket_mem - cur_socket_mem; + + /* make it multiple of page size */ + max_pagesz_mem = RTE_ALIGN_FLOOR(max_pagesz_mem, + hugepage_sz); + + RTE_LOG(DEBUG, EAL, "Attempting to preallocate " + "%" PRIu64 "M on socket %i\n", + max_pagesz_mem >> 20, socket_id); + + type_msl_idx = 0; + while (cur_pagesz_mem < max_pagesz_mem && + total_segs < max_segs) { + uint64_t cur_mem; + unsigned int n_segs; + + if (msl_idx >= RTE_MAX_MEMSEG_LISTS) { + RTE_LOG(ERR, EAL, + "No more space in memseg lists, please increase %s\n", + RTE_STR(CONFIG_RTE_MAX_MEMSEG_LISTS)); + return -1; + } + + msl = &mcfg->memsegs[msl_idx]; + + cur_mem = get_mem_amount(hugepage_sz, + max_pagesz_mem); + n_segs = cur_mem / hugepage_sz; + + if (alloc_memseg_list(msl, hugepage_sz, n_segs, + socket_id, type_msl_idx)) { + /* failing to allocate a memseg list is + * a serious error. + */ + RTE_LOG(ERR, EAL, "Cannot allocate memseg list\n"); + return -1; + } + + if (alloc_va_space(msl)) { + /* if we couldn't allocate VA space, we + * can try with smaller page sizes. + */ + RTE_LOG(ERR, EAL, "Cannot allocate VA space for memseg list, retrying with different page size\n"); + /* deallocate memseg list */ + if (free_memseg_list(msl)) + return -1; + break; + } + + total_segs += msl->memseg_arr.len; + cur_pagesz_mem = total_segs * hugepage_sz; + type_msl_idx++; + msl_idx++; + } + cur_socket_mem += cur_pagesz_mem; + } + if (cur_socket_mem == 0) { + RTE_LOG(ERR, EAL, "Cannot allocate VA space on socket %u\n", + socket_id); + return -1; + } + } + + return 0; +} + +static int __rte_unused +memseg_primary_init(void) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + int i, socket_id, hpi_idx, msl_idx = 0; + struct rte_memseg_list *msl; + uint64_t max_mem, total_mem; + + /* no-huge does not need this at all */ + if (internal_config.no_hugetlbfs) + return 0; + + max_mem = (uint64_t)RTE_MAX_MEM_MB << 20; + total_mem = 0; + + /* create memseg lists */ + for (hpi_idx = 0; hpi_idx < (int) internal_config.num_hugepage_sizes; + hpi_idx++) { + struct hugepage_info *hpi; + uint64_t hugepage_sz; + + hpi = &internal_config.hugepage_info[hpi_idx]; + hugepage_sz = hpi->hugepage_sz; + + for (i = 0; i < (int) rte_socket_count(); i++) { + uint64_t max_type_mem, total_type_mem = 0; + int type_msl_idx, max_segs, total_segs = 0; + + socket_id = rte_socket_id_by_idx(i); + +#ifndef RTE_EAL_NUMA_AWARE_HUGEPAGES + if (socket_id > 0) + break; +#endif + + if (total_mem >= max_mem) + break; + + max_type_mem = RTE_MIN(max_mem - total_mem, + (uint64_t)RTE_MAX_MEM_MB_PER_TYPE << 20); + max_segs = RTE_MAX_MEMSEG_PER_TYPE; + + type_msl_idx = 0; + while (total_type_mem < max_type_mem && + total_segs < max_segs) { + uint64_t cur_max_mem, cur_mem; + unsigned int n_segs; + + if (msl_idx >= RTE_MAX_MEMSEG_LISTS) { + RTE_LOG(ERR, EAL, + "No more space in memseg lists, please increase %s\n", + RTE_STR(CONFIG_RTE_MAX_MEMSEG_LISTS)); + return -1; + } + + msl = &mcfg->memsegs[msl_idx++]; + + cur_max_mem = max_type_mem - total_type_mem; + + cur_mem = get_mem_amount(hugepage_sz, + cur_max_mem); + n_segs = cur_mem / hugepage_sz; + + if (alloc_memseg_list(msl, hugepage_sz, n_segs, + socket_id, type_msl_idx)) + return -1; + + total_segs += msl->memseg_arr.len; + total_type_mem = total_segs * hugepage_sz; + type_msl_idx++; + + if (alloc_va_space(msl)) { + RTE_LOG(ERR, EAL, "Cannot allocate VA space for memseg list\n"); + return -1; + } + } + total_mem += total_type_mem; + } + } + return 0; +} + +static int +memseg_secondary_init(void) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + int msl_idx = 0; + struct rte_memseg_list *msl; + + for (msl_idx = 0; msl_idx < RTE_MAX_MEMSEG_LISTS; msl_idx++) { + + msl = &mcfg->memsegs[msl_idx]; + + /* skip empty memseg lists */ + if (msl->memseg_arr.len == 0) + continue; + + if (rte_fbarray_attach(&msl->memseg_arr)) { + RTE_LOG(ERR, EAL, "Cannot attach to primary process memseg lists\n"); + return -1; + } + + /* preallocate VA space */ + if (alloc_va_space(msl)) { + RTE_LOG(ERR, EAL, "Cannot preallocate VA space for hugepage memory\n"); + return -1; + } + } + + return 0; +} + +int +rte_eal_memseg_init(void) +{ + return rte_eal_process_type() == RTE_PROC_PRIMARY ? +#ifndef RTE_ARCH_64 + memseg_primary_init_32() : +#else + memseg_primary_init() : +#endif + memseg_secondary_init(); +} |