X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=lib%2Flibrte_eal%2Flinuxapp%2Feal%2Feal_memory.c;h=daab36490d46c61b7c706bd6e02300ec5f1c188b;hb=2a04139f66b457d069445a65fd11722d91463bcb;hp=3a1822ee2e848e8e86d57ce94de9c9c8045f9d26;hpb=f622f3848fb17400367781895e3b9c1a51e0a43d;p=dpdk.git diff --git a/lib/librte_eal/linuxapp/eal/eal_memory.c b/lib/librte_eal/linuxapp/eal/eal_memory.c index 3a1822ee2e..daab36490d 100644 --- a/lib/librte_eal/linuxapp/eal/eal_memory.c +++ b/lib/librte_eal/linuxapp/eal/eal_memory.c @@ -1,74 +1,17 @@ -/*- - * BSD LICENSE - * - * Copyright(c) 2010-2014 Intel Corporation. All rights reserved. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Intel Corporation nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ -/* BSD LICENSE - * - * Copyright(c) 2013 6WIND. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions - * are met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above copyright - * notice, this list of conditions and the following disclaimer in - * the documentation and/or other materials provided with the - * distribution. - * * Neither the name of 6WIND S.A. nor the names of its - * contributors may be used to endorse or promote products derived - * from this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2010-2014 Intel Corporation. + * Copyright(c) 2013 6WIND S.A. */ +#define _FILE_OFFSET_BITS 64 #include #include +#include #include #include #include #include #include -#include #include #include #include @@ -76,15 +19,19 @@ #include #include #include -#include #include #include +#include +#include +#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES +#include +#include +#endif +#include #include #include -#include #include -#include #include #include #include @@ -97,6 +44,8 @@ #include "eal_filesystem.h" #include "eal_hugepages.h" +#define PFN_MASK_SIZE 8 + /** * @file * Huge page mapping under linux @@ -109,10 +58,138 @@ * 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" +static void +test_phys_addrs_available(void) +{ + uint64_t tmp; + phys_addr_t physaddr; + + if (!rte_eal_has_hugepages()) { + RTE_LOG(ERR, EAL, + "Started without hugepages support, physical addresses not available\n"); + phys_addrs_available = false; + return; + } + + physaddr = rte_mem_virt2phy(&tmp); + if (physaddr == RTE_BAD_PHYS_ADDR) { + if (rte_eal_iova_mode() == RTE_IOVA_PA) + RTE_LOG(ERR, EAL, + "Cannot obtain physical addresses: %s. " + "Only vfio will function.\n", + strerror(errno)); + phys_addrs_available = false; + } +} + +/* + * Get physical address of any mapped virtual address in the current process. + */ +phys_addr_t +rte_mem_virt2phy(const void *virtaddr) +{ + int fd, retval; + uint64_t page, physaddr; + unsigned long virt_pfn; + int page_size; + off_t offset; + + /* Cannot parse /proc/self/pagemap, no need to log errors everywhere */ + if (!phys_addrs_available) + return RTE_BAD_IOVA; + + /* standard page size */ + page_size = getpagesize(); + + fd = open("/proc/self/pagemap", O_RDONLY); + if (fd < 0) { + RTE_LOG(ERR, EAL, "%s(): cannot open /proc/self/pagemap: %s\n", + __func__, strerror(errno)); + return RTE_BAD_IOVA; + } + + virt_pfn = (unsigned long)virtaddr / page_size; + offset = sizeof(uint64_t) * virt_pfn; + if (lseek(fd, offset, SEEK_SET) == (off_t) -1) { + RTE_LOG(ERR, EAL, "%s(): seek error in /proc/self/pagemap: %s\n", + __func__, strerror(errno)); + close(fd); + return RTE_BAD_IOVA; + } + + retval = read(fd, &page, PFN_MASK_SIZE); + close(fd); + if (retval < 0) { + RTE_LOG(ERR, EAL, "%s(): cannot read /proc/self/pagemap: %s\n", + __func__, strerror(errno)); + return RTE_BAD_IOVA; + } else if (retval != PFN_MASK_SIZE) { + RTE_LOG(ERR, EAL, "%s(): read %d bytes from /proc/self/pagemap " + "but expected %d:\n", + __func__, retval, PFN_MASK_SIZE); + return RTE_BAD_IOVA; + } + + /* + * the pfn (page frame number) are bits 0-54 (see + * pagemap.txt in linux Documentation) + */ + if ((page & 0x7fffffffffffffULL) == 0) + return RTE_BAD_IOVA; + + physaddr = ((page & 0x7fffffffffffffULL) * page_size) + + ((unsigned long)virtaddr % page_size); + + return physaddr; +} + +rte_iova_t +rte_mem_virt2iova(const void *virtaddr) +{ + if (rte_eal_iova_mode() == RTE_IOVA_VA) + return (uintptr_t)virtaddr; + return rte_mem_virt2phy(virtaddr); +} + +/* + * For each hugepage in hugepg_tbl, fill the physaddr value. We find + * it by browsing the /proc/self/pagemap special file. + */ +static int +find_physaddrs(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi) +{ + unsigned int i; + phys_addr_t addr; + + for (i = 0; i < hpi->num_pages[0]; i++) { + addr = rte_mem_virt2phy(hugepg_tbl[i].orig_va); + if (addr == RTE_BAD_PHYS_ADDR) + return -1; + hugepg_tbl[i].physaddr = addr; + } + return 0; +} + +/* + * For each hugepage in hugepg_tbl, fill the physaddr value sequentially. + */ +static int +set_physaddrs(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi) +{ + unsigned int i; + static phys_addr_t addr; + + for (i = 0; i < hpi->num_pages[0]; i++) { + hugepg_tbl[i].physaddr = addr; + addr += hugepg_tbl[i].size; + } + return 0; +} + /* * Check whether address-space layout randomization is enabled in * the kernel. This is important for multi-process as it can prevent @@ -143,228 +220,196 @@ aslr_enabled(void) } } -/* - * Try to mmap *size bytes in /dev/zero. If it is succesful, 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; - int fd; - long aligned_addr; - - if (internal_config.base_virtaddr != 0) { - addr = (void*) (uintptr_t) (internal_config.base_virtaddr + - baseaddr_offset); - } - else addr = NULL; - - RTE_LOG(INFO, EAL, "Ask a virtual area of 0x%zx bytes\n", *size); +static sigjmp_buf huge_jmpenv; - fd = open("/dev/zero", O_RDONLY); - if (fd < 0){ - RTE_LOG(ERR, EAL, "Cannot open /dev/zero\n"); - return NULL; - } - do { - addr = mmap(addr, - (*size) + hugepage_sz, PROT_READ, MAP_PRIVATE, fd, 0); - if (addr == MAP_FAILED) - *size -= hugepage_sz; - } while (addr == MAP_FAILED && *size > 0); - - if (addr == MAP_FAILED) { - close(fd); - RTE_LOG(INFO, EAL, "Cannot get a virtual area\n"); - 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(INFO, EAL, "Virtual area found at %p (size = 0x%zx)\n", - addr, *size); +static void huge_sigbus_handler(int signo __rte_unused) +{ + siglongjmp(huge_jmpenv, 1); +} - /* increment offset */ - baseaddr_offset += *size; +/* Put setjmp into a wrap method to avoid compiling error. Any non-volatile, + * non-static local variable in the stack frame calling sigsetjmp might be + * clobbered by a call to longjmp. + */ +static int huge_wrap_sigsetjmp(void) +{ + return sigsetjmp(huge_jmpenv, 1); +} - return addr; +#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES +/* Callback for numa library. */ +void numa_error(char *where) +{ + RTE_LOG(ERR, EAL, "%s failed: %s\n", where, strerror(errno)); } +#endif /* * Mmap all hugepages of hugepage table: it first open a file in * hugetlbfs, then mmap() hugepage_sz data in it. If orig is set, the * virtual address is stored in hugepg_tbl[i].orig_va, else it is stored * in hugepg_tbl[i].final_va. The second mapping (when orig is 0) tries to - * map continguous physical blocks in contiguous virtual blocks. + * map contiguous physical blocks in contiguous virtual blocks. */ -static int -map_all_hugepages(struct hugepage *hugepg_tbl, - struct hugepage_info *hpi, int orig) +static unsigned +map_all_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, + uint64_t *essential_memory __rte_unused) { int fd; unsigned i; void *virtaddr; - void *vma_addr = NULL; - size_t vma_len = 0; + struct flock lck = {0}; +#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES + int node_id = -1; + int essential_prev = 0; + int oldpolicy; + struct bitmask *oldmask = numa_allocate_nodemask(); + bool have_numa = true; + unsigned long maxnode = 0; + + /* Check if kernel supports NUMA. */ + if (numa_available() != 0) { + RTE_LOG(DEBUG, EAL, "NUMA is not supported.\n"); + have_numa = false; + } - for (i = 0; i < hpi->num_pages[0]; i++) { - size_t hugepage_sz = hpi->hugepage_sz; - - 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_X86_64 - /* for 32-bit systems, don't remap 1G pages, just reuse original - * map address as final map address. - */ - else if (hugepage_sz == RTE_PGSIZE_1G){ - hugepg_tbl[i].final_va = hugepg_tbl[i].orig_va; - hugepg_tbl[i].orig_va = NULL; - continue; + 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) { + RTE_LOG(ERR, EAL, + "Failed to get current mempolicy: %s. " + "Assuming MPOL_DEFAULT.\n", strerror(errno)); + oldpolicy = MPOL_DEFAULT; } + for (i = 0; i < RTE_MAX_NUMA_NODES; i++) + if (internal_config.socket_mem[i]) + maxnode = i + 1; + } #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++) { - if (hugepg_tbl[j].physaddr != - hugepg_tbl[j-1].physaddr + hugepage_sz) + + 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 + if (maxnode) { + unsigned int j; + + for (j = 0; j < maxnode; j++) + if (essential_memory[j]) break; + + if (j == maxnode) { + node_id = (node_id + 1) % maxnode; + while (!internal_config.socket_mem[node_id]) { + node_id++; + node_id %= maxnode; + } + essential_prev = 0; + } else { + node_id = j; + essential_prev = essential_memory[j]; + + if (essential_memory[j] < hugepage_sz) + essential_memory[j] = 0; + else + essential_memory[j] -= hugepage_sz; } - 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; + + RTE_LOG(DEBUG, EAL, + "Setting policy MPOL_PREFERRED for socket %d\n", + node_id); + numa_set_preferred(node_id); } +#endif + + 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, 0755); + fd = open(hf->filepath, O_CREAT | O_RDWR, 0600); if (fd < 0) { - RTE_LOG(ERR, EAL, "%s(): open failed: %s\n", __func__, + RTE_LOG(DEBUG, EAL, "%s(): open failed: %s\n", __func__, strerror(errno)); - return -1; + goto out; } - virtaddr = mmap(vma_addr, hugepage_sz, PROT_READ | PROT_WRITE, - MAP_SHARED, fd, 0); + /* map the segment, and populate page tables, + * 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(ERR, EAL, "%s(): mmap failed: %s\n", __func__, + RTE_LOG(DEBUG, EAL, "%s(): mmap failed: %s\n", __func__, strerror(errno)); close(fd); - return -1; + goto out; } - if (orig) { - hugepg_tbl[i].orig_va = virtaddr; - memset(virtaddr, 0, hugepage_sz); - } - else { - hugepg_tbl[i].final_va = virtaddr; - } + hf->orig_va = virtaddr; - /* set shared flock on the file. */ - if (flock(fd, LOCK_SH | LOCK_NB) == -1) { - RTE_LOG(ERR, EAL, "%s(): Locking file failed:%s \n", - __func__, strerror(errno)); + /* 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); - return -1; + unlink(hugepg_tbl[i].filepath); +#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES + if (maxnode) + essential_memory[node_id] = + essential_prev; +#endif + goto out; } + *(int *)virtaddr = 0; - close(fd); - - vma_addr = (char *)vma_addr + hugepage_sz; - vma_len -= hugepage_sz; - } - return 0; -} -/* Unmap all hugepages from original mapping. */ -static int -unmap_all_hugepages_orig(struct hugepage *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; + /* set shared lock on the file. */ + lck.l_type = F_RDLCK; + lck.l_whence = SEEK_SET; + lck.l_start = 0; + lck.l_len = hugepage_sz; + if (fcntl(fd, F_SETLK, &lck) == -1) { + RTE_LOG(DEBUG, EAL, "%s(): Locking file failed:%s \n", + __func__, strerror(errno)); + close(fd); + goto out; } - } - return 0; -} - -/* - * For each hugepage in hugepg_tbl, fill the physaddr value. We find - * it by browsing the /proc/self/pagemap special file. - */ -static int -find_physaddr(struct hugepage *hugepg_tbl, struct hugepage_info *hpi) -{ - int fd; - unsigned i; - uint64_t page; - unsigned long virt_pfn; - int page_size; - - /* standard page size */ - page_size = getpagesize(); - fd = open("/proc/self/pagemap", O_RDONLY); - if (fd < 0) { - RTE_LOG(ERR, EAL, "%s(): cannot open /proc/self/pagemap: %s\n", - __func__, strerror(errno)); - return -1; + close(fd); } - for (i = 0; i < hpi->num_pages[0]; i++) { - off_t offset; - virt_pfn = (unsigned long)hugepg_tbl[i].orig_va / - page_size; - offset = sizeof(uint64_t) * virt_pfn; - if (lseek(fd, offset, SEEK_SET) == (off_t) -1) { - RTE_LOG(ERR, EAL, "%s(): seek error in /proc/self/pagemap: %s\n", - __func__, strerror(errno)); - close(fd); - return -1; - } - if (read(fd, &page, sizeof(uint64_t)) < 0) { - RTE_LOG(ERR, EAL, "%s(): cannot read /proc/self/pagemap: %s\n", - __func__, strerror(errno)); - close(fd); - return -1; +out: +#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES + if (maxnode) { + RTE_LOG(DEBUG, EAL, + "Restoring previous memory policy: %d\n", oldpolicy); + if (oldpolicy == MPOL_DEFAULT) { + numa_set_localalloc(); + } else if (set_mempolicy(oldpolicy, oldmask->maskp, + oldmask->size + 1) < 0) { + RTE_LOG(ERR, EAL, "Failed to restore mempolicy: %s\n", + strerror(errno)); + numa_set_localalloc(); } - - /* - * the pfn (page frame number) are bits 0-54 (see - * pagemap.txt in linux Documentation) - */ - hugepg_tbl[i].physaddr = ((page & 0x7fffffffffffffULL) * page_size); } - close(fd); - return 0; + numa_free_cpumask(oldmask); +#endif + return i; } /* @@ -372,7 +417,7 @@ find_physaddr(struct hugepage *hugepg_tbl, struct hugepage_info *hpi) * page. */ static int -find_numasocket(struct hugepage *hugepg_tbl, struct hugepage_info *hpi) +find_numasocket(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi) { int socket_id; char *end, *nodestr; @@ -384,13 +429,13 @@ find_numasocket(struct hugepage *hugepg_tbl, struct hugepage_info *hpi) f = fopen("/proc/self/numa_maps", "r"); if (f == NULL) { - RTE_LOG(INFO, EAL, "cannot open /proc/self/numa_maps," - " consider that all memory is in socket_id 0\n"); + RTE_LOG(NOTICE, EAL, "NUMA support not available" + " consider that all memory is in socket_id 0\n"); return 0; } - rte_snprintf(hugedir_str, sizeof(hugedir_str), - "%s/", hpi->hugedir); + snprintf(hugedir_str, sizeof(hugedir_str), + "%s/%s", hpi->hugedir, internal_config.hugefile_prefix); /* parse numa map */ while (fgets(buf, sizeof(buf), f) != NULL) { @@ -434,6 +479,11 @@ find_numasocket(struct hugepage *hugepg_tbl, struct hugepage_info *hpi) if (hugepg_tbl[i].orig_va == va) { hugepg_tbl[i].socket_id = socket_id; hp_count++; +#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES + RTE_LOG(DEBUG, EAL, + "Hugepage %s is on socket %d\n", + hugepg_tbl[i].filepath, socket_id); +#endif } } } @@ -449,49 +499,23 @@ error: return -1; } -/* - * Sort the hugepg_tbl by physical address (lower addresses first). We - * use a slow algorithm, but we won't have millions of pages, and this - * is only done at init time. - */ static int -sort_by_physaddr(struct hugepage *hugepg_tbl, struct hugepage_info *hpi) +cmp_physaddr(const void *a, const void *b) { - unsigned i, j; - int smallest_idx; - uint64_t smallest_addr; - struct hugepage tmp; - - for (i = 0; i < hpi->num_pages[0]; i++) { - smallest_addr = 0; - smallest_idx = -1; - - /* - * browse all entries starting at 'i', and find the - * entry with the smallest addr - */ - for (j=i; j< hpi->num_pages[0]; j++) { - - if (smallest_addr == 0 || - hugepg_tbl[j].physaddr < smallest_addr) { - smallest_addr = hugepg_tbl[j].physaddr; - smallest_idx = j; - } - } - - /* should not happen */ - if (smallest_idx == -1) { - RTE_LOG(ERR, EAL, "%s(): error in physaddr sorting\n", __func__); - return -1; - } - - /* swap the 2 entries in the table */ - memcpy(&tmp, &hugepg_tbl[smallest_idx], sizeof(struct hugepage)); - memcpy(&hugepg_tbl[smallest_idx], &hugepg_tbl[i], - sizeof(struct hugepage)); - memcpy(&hugepg_tbl[i], &tmp, sizeof(struct hugepage)); - } - return 0; +#ifndef RTE_ARCH_PPC_64 + const struct hugepage_file *p1 = a; + const struct hugepage_file *p2 = b; +#else + /* PowerPC needs memory sorted in reverse order from x86 */ + const struct hugepage_file *p1 = b; + const struct hugepage_file *p2 = a; +#endif + if (p1->physaddr < p2->physaddr) + return -1; + else if (p1->physaddr > p2->physaddr) + return 1; + else + return 0; } /* @@ -511,6 +535,8 @@ create_shared_memory(const char *filename, const size_t mem_size) } retval = mmap(NULL, mem_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); close(fd); + if (retval == MAP_FAILED) + return NULL; return retval; } @@ -519,29 +545,53 @@ create_shared_memory(const char *filename, const size_t mem_size) * destination is typically the shared memory. */ static int -copy_hugepages_to_shared_mem(struct hugepage * dst, int dest_size, - const struct hugepage * src, int src_size) +copy_hugepages_to_shared_mem(struct hugepage_file * dst, int dest_size, + const struct hugepage_file * src, int src_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; - memcpy(&dst[dst_pos], &src[src_pos], sizeof(struct hugepage)); + memcpy(&dst[dst_pos], &src[src_pos], sizeof(struct hugepage_file)); dst_pos++; } } return 0; } +static int +unlink_hugepage_files(struct hugepage_file *hugepg_tbl, + unsigned num_hp_info) +{ + unsigned socket, size; + int page, nrpages = 0; + + /* get total number of hugepages */ + for (size = 0; size < num_hp_info; size++) + for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) + nrpages += + internal_config.hugepage_info[size].num_pages[socket]; + + for (page = 0; page < nrpages; page++) { + struct hugepage_file *hp = &hugepg_tbl[page]; + + if (hp->final_va != NULL && unlink(hp->filepath)) { + RTE_LOG(WARNING, EAL, "%s(): Removing %s failed: %s\n", + __func__, hp->filepath, strerror(errno)); + } + } + return 0; +} + /* * unmaps hugepages that are not going to be used. since we originally allocate * ALL hugepages (not just those we need), additional unmapping needs to be done. */ static int -unmap_unneeded_hugepages(struct hugepage *hugepg_tbl, +unmap_unneeded_hugepages(struct hugepage_file *hugepg_tbl, struct hugepage_info *hpi, unsigned num_hp_info) { @@ -556,9 +606,10 @@ unmap_unneeded_hugepages(struct hugepage *hugepg_tbl, for (size = 0; size < num_hp_info; size++) { for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) { unsigned pages_found = 0; + /* traverse until we have unmapped all the unused pages */ for (page = 0; page < nrpages; page++) { - struct hugepage *hp = &hugepg_tbl[page]; + struct hugepage_file *hp = &hugepg_tbl[page]; /* find a page that matches the criteria */ if ((hp->size == hpi[size].hugepage_sz) && @@ -566,17 +617,24 @@ unmap_unneeded_hugepages(struct hugepage *hugepg_tbl, /* if we skipped enough pages, unmap the rest */ if (pages_found == hpi[size].num_pages[socket]) { - munmap(hp->final_va, hp->size); - hp->final_va = NULL; - if (remove(hp->filepath) == -1) { + uint64_t unmap_len; + + unmap_len = hp->size; + + /* get start addr and len of the remaining segment */ + munmap(hp->orig_va, + (size_t)unmap_len); + + hp->orig_va = NULL; + if (unlink(hp->filepath) == -1) { RTE_LOG(ERR, EAL, "%s(): Removing %s failed: %s\n", __func__, hp->filepath, strerror(errno)); return -1; } - } - /* lock the page and skip */ - else + } else { + /* lock the page and skip */ pages_found++; + } } /* match page */ } /* foreach page */ @@ -586,6 +644,413 @@ unmap_unneeded_hugepages(struct hugepage *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); + struct flock lck; + 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. */ + lck.l_type = F_RDLCK; + lck.l_whence = SEEK_SET; + lck.l_start = 0; + lck.l_len = page_sz; + if (fcntl(fd, F_SETLK, &lck) == -1) { + 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; +} + +#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) { @@ -598,7 +1063,7 @@ get_socket_mem_size(int socket) size += hpi->hugepage_sz * hpi->num_pages[socket]; } - return (size); + return size; } /* @@ -622,13 +1087,76 @@ calc_num_pages_per_socket(uint64_t * memory, if (num_hp_info == 0) return -1; - for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_mem != 0; socket++) { - /* if specific memory amounts per socket weren't requested */ - if (internal_config.force_sockets == 0) { + /* 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; + unsigned lcore_id; + + /* Compute number of cores per socket */ + memset(cpu_per_socket, 0, sizeof(cpu_per_socket)); + RTE_LCORE_FOREACH(lcore_id) { + cpu_per_socket[rte_lcore_to_socket_id(lcore_id)]++; + } + + /* + * Automatically spread requested memory amongst detected sockets according + * to number of cores from cpu mask present on each socket + */ + total_size = internal_config.memory; + for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_size != 0; socket++) { + + /* Set memory amount per socket */ + default_size = (internal_config.memory * cpu_per_socket[socket]) + / rte_lcore_count(); + + /* Limit to maximum available memory on socket */ + default_size = RTE_MIN(default_size, get_socket_mem_size(socket)); + + /* Update sizes */ + memory[socket] = default_size; + total_size -= default_size; + } + + /* + * If some memory is remaining, try to allocate it by getting all + * available memory from sockets, one after the other + */ + for (socket = 0; socket < RTE_MAX_NUMA_NODES && total_size != 0; socket++) { /* take whatever is available */ - memory[socket] = RTE_MIN(get_socket_mem_size(socket), - total_mem); + default_size = RTE_MIN(get_socket_mem_size(socket) - memory[socket], + 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; @@ -673,13 +1201,14 @@ 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); available = requested - ((unsigned) (memory[socket] / 0x100000)); - RTE_LOG(INFO, EAL, "Not enough memory available on socket %u! " + RTE_LOG(ERR, EAL, "Not enough memory available on socket %u! " "Requested: %uMB, available: %uMB\n", socket, requested, available); return -1; @@ -690,13 +1219,58 @@ calc_num_pages_per_socket(uint64_t * memory, if (total_mem > 0) { requested = (unsigned) (internal_config.memory / 0x100000); available = requested - (unsigned) (total_mem / 0x100000); - RTE_LOG(INFO, EAL, "Not enough memory available! Requested: %uMB," + RTE_LOG(ERR, EAL, "Not enough memory available! Requested: %uMB," " available: %uMB\n", requested, available); return -1; } return total_num_pages; } +static inline size_t +eal_get_hugepage_mem_size(void) +{ + uint64_t size = 0; + unsigned i, j; + + for (i = 0; i < internal_config.num_hugepage_sizes; i++) { + struct hugepage_info *hpi = &internal_config.hugepage_info[i]; + if (hpi->hugedir != NULL) { + for (j = 0; j < RTE_MAX_NUMA_NODES; j++) { + size += hpi->hugepage_sz * hpi->num_pages[j]; + } + } + } + + return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX; +} + +static struct sigaction huge_action_old; +static int huge_need_recover; + +static void +huge_register_sigbus(void) +{ + sigset_t mask; + struct sigaction action; + + sigemptyset(&mask); + sigaddset(&mask, SIGBUS); + action.sa_flags = 0; + action.sa_mask = mask; + action.sa_handler = huge_sigbus_handler; + + huge_need_recover = !sigaction(SIGBUS, &action, &huge_action_old); +} + +static void +huge_recover_sigbus(void) +{ + if (huge_need_recover) { + sigaction(SIGBUS, &huge_action_old, NULL); + huge_need_recover = 0; + } +} + /* * Prepare physical memory mapping: fill configuration structure with * these infos, return 0 on success. @@ -709,42 +1283,88 @@ calc_num_pages_per_socket(uint64_t * memory, * 7. fill memsegs in configuration with contiguous zones */ static int -rte_eal_hugepage_init(void) +eal_legacy_hugepage_init(void) { struct rte_mem_config *mcfg; - struct hugepage *hugepage, *tmp_hp = NULL; + 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 nrpages, total_pages = 0; + int i, j; + int nr_hugefiles, nr_hugepages = 0; void *addr; + test_phys_addrs_available(); + memset(used_hp, 0, sizeof(used_hp)); /* get pointer to global configuration */ mcfg = rte_eal_get_configuration()->mem_config; - /* for debug purposes, hugetlbfs can be disabled */ + /* hugetlbfs can be disabled */ if (internal_config.no_hugetlbfs) { - addr = malloc(internal_config.memory); - mcfg->memseg[0].phys_addr = (phys_addr_t)(uintptr_t)addr; - mcfg->memseg[0].addr = addr; - mcfg->memseg[0].len = internal_config.memory; - mcfg->memseg[0].socket_id = 0; + 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); + if (addr == MAP_FAILED) { + RTE_LOG(ERR, EAL, "%s: mmap() failed: %s\n", __func__, + strerror(errno)); + return -1; + } + 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; } - /* calculate total number of hugepages available. at this point we haven't * yet started sorting them so they all are on socket 0 */ for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++) { /* meanwhile, also initialize used_hp hugepage sizes in used_hp */ used_hp[i].hugepage_sz = internal_config.hugepage_info[i].hugepage_sz; - total_pages += internal_config.hugepage_info[i].num_pages[0]; + nr_hugepages += internal_config.hugepage_info[i].num_pages[0]; } /* @@ -753,16 +1373,23 @@ rte_eal_hugepage_init(void) * processing done on these pages, shared memory will be created * at a later stage. */ - tmp_hp = malloc(total_pages * sizeof(struct hugepage)); + tmp_hp = malloc(nr_hugepages * sizeof(struct hugepage_file)); if (tmp_hp == NULL) goto fail; - memset(tmp_hp, 0, total_pages * sizeof(struct hugepage)); + memset(tmp_hp, 0, nr_hugepages * sizeof(struct hugepage_file)); hp_offset = 0; /* where we start the current page size entries */ + huge_register_sigbus(); + + /* make a copy of socket_mem, needed for balanced allocation. */ + 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; struct hugepage_info *hpi; /* @@ -772,21 +1399,43 @@ rte_eal_hugepage_init(void) */ hpi = &internal_config.hugepage_info[i]; - if (hpi->num_pages == 0) + if (hpi->num_pages[0] == 0) continue; /* map all hugepages available */ - if (map_all_hugepages(&tmp_hp[hp_offset], hpi, 1) < 0){ - RTE_LOG(DEBUG, EAL, "Failed to mmap %u MB hugepages\n", - (unsigned)(hpi->hugepage_sz / 0x100000)); - goto fail; + pages_old = hpi->num_pages[0]; + 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", + pages_new, pages_old, + (unsigned)(hpi->hugepage_sz / 0x100000)); + + int pages = pages_old - pages_new; + + nr_hugepages -= pages; + hpi->num_pages[0] = pages_new; + if (pages_new == 0) + continue; } - /* find physical addresses and sockets for each hugepage */ - if (find_physaddr(&tmp_hp[hp_offset], hpi) < 0){ - RTE_LOG(DEBUG, EAL, "Failed to find phys addr for %u MB pages\n", - (unsigned)(hpi->hugepage_sz / 0x100000)); - goto fail; + 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 " + "for %u MB pages\n", + (unsigned int)(hpi->hugepage_sz / 0x100000)); + goto fail; + } + } else { + /* set physical addresses for each hugepage */ + if (set_physaddrs(&tmp_hp[hp_offset], hpi) < 0) { + RTE_LOG(DEBUG, EAL, "Failed to set phys addr " + "for %u MB pages\n", + (unsigned int)(hpi->hugepage_sz / 0x100000)); + goto fail; + } } if (find_numasocket(&tmp_hp[hp_offset], hpi) < 0){ @@ -795,35 +1444,34 @@ rte_eal_hugepage_init(void) goto fail; } - if (sort_by_physaddr(&tmp_hp[hp_offset], hpi) < 0) - goto fail; - - /* remap all hugepages */ - if (map_all_hugepages(&tmp_hp[hp_offset], hpi, 0) < 0){ - RTE_LOG(DEBUG, 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; + qsort(&tmp_hp[hp_offset], hpi->num_pages[0], + sizeof(struct hugepage_file), cmp_physaddr); /* we have processed a num of hugepages of this size, so inc offset */ hp_offset += hpi->num_pages[0]; } + huge_recover_sigbus(); + + if (internal_config.memory == 0 && internal_config.force_sockets == 0) + internal_config.memory = eal_get_hugepage_mem_size(); + + nr_hugefiles = nr_hugepages; + + /* clean out the numbers of pages */ for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++) for (j = 0; j < RTE_MAX_NUMA_NODES; j++) internal_config.hugepage_info[i].num_pages[j] = 0; /* get hugepages for each socket */ - for (i = 0; i < total_pages; i++) { + for (i = 0; i < nr_hugefiles; i++) { int socket = tmp_hp[i].socket_id; /* find a hugepage info with right size and increment num_pages */ - for (j = 0; j < (int) internal_config.num_hugepage_sizes; j++) { + const int nb_hpsizes = RTE_MIN(MAX_HUGEPAGE_SIZES, + (int)internal_config.num_hugepage_sizes); + for (j = 0; j < nb_hpsizes; j++) { if (tmp_hp[i].size == internal_config.hugepage_info[j].hugepage_sz) { internal_config.hugepage_info[j].num_pages[socket]++; @@ -836,37 +1484,38 @@ rte_eal_hugepage_init(void) memory[i] = internal_config.socket_mem[i]; /* calculate final number of pages */ - nrpages = calc_num_pages_per_socket(memory, + nr_hugepages = calc_num_pages_per_socket(memory, internal_config.hugepage_info, used_hp, internal_config.num_hugepage_sizes); /* error if not enough memory available */ - if (nrpages < 0) + if (nr_hugepages < 0) goto fail; /* reporting in! */ for (i = 0; i < (int) internal_config.num_hugepage_sizes; i++) { for (j = 0; j < RTE_MAX_NUMA_NODES; j++) { if (used_hp[i].num_pages[j] > 0) { - RTE_LOG(INFO, EAL, - "Requesting %u pages of size %uMB" - " from socket %i\n", - used_hp[i].num_pages[j], - (unsigned) - (used_hp[i].hugepage_sz / 0x100000), - j); + RTE_LOG(DEBUG, EAL, + "Requesting %u pages of size %uMB" + " from socket %i\n", + used_hp[i].num_pages[j], + (unsigned) + (used_hp[i].hugepage_sz / 0x100000), + j); } } } /* create shared memory */ hugepage = create_shared_memory(eal_hugepage_info_path(), - nrpages * sizeof(struct hugepage)); + nr_hugefiles * sizeof(struct hugepage_file)); if (hugepage == NULL) { RTE_LOG(ERR, EAL, "Failed to create shared memory!\n"); goto fail; } + memset(hugepage, 0, nr_hugefiles * sizeof(struct hugepage_file)); /* * unmap pages that we won't need (looks at used_hp). @@ -880,74 +1529,75 @@ 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, nrpages, - tmp_hp, total_pages) < 0) { + if (copy_hugepages_to_shared_mem(hugepage, nr_hugefiles, + tmp_hp, nr_hugefiles) < 0) { RTE_LOG(ERR, EAL, "Copying tables to shared memory failed!\n"); 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) { + RTE_LOG(ERR, EAL, "Unlinking hugepage files failed!\n"); + goto fail; + } + /* free the temporary hugepage table */ free(tmp_hp); tmp_hp = NULL; - memset(mcfg->memseg, 0, sizeof(mcfg->memseg)); - j = -1; - for (i = 0; i < nrpages; 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; - 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; + 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].phys_addr = 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 { - 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 < nrpages) { - 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, nrpages, RTE_STR(CONFIG_RTE_MAX_MEMSEG), - RTE_MAX_MEMSEG); - return (-ENOMEM); + /* destroy backing fbarray */ + rte_fbarray_destroy(&msl->memseg_arr); } - return 0; - fail: - if (tmp_hp) - free(tmp_hp); + huge_recover_sigbus(); + free(tmp_hp); + if (hugepage != NULL) + munmap(hugepage, nr_hugefiles * sizeof(struct hugepage_file)); + return -1; } @@ -970,14 +1620,14 @@ getFileSize(int fd) * in order to form a contiguous block in the virtual memory space */ static int -rte_eal_hugepage_attach(void) +eal_legacy_hugepage_attach(void) { - const struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; - const struct hugepage *hp = NULL; - unsigned num_hp = 0; - unsigned i, s = 0; /* s used to track the segment number */ - off_t size; - int fd, fd_zero = -1, fd_hugepage = -1; + struct hugepage_file *hp = NULL; + unsigned int num_hp = 0; + unsigned int i = 0; + unsigned int cur_seg; + off_t size = 0; + int fd, fd_hugepage = -1; if (aslr_enabled() > 0) { RTE_LOG(WARNING, EAL, "WARNING: Address Space Layout Randomization " @@ -986,143 +1636,108 @@ rte_eal_hugepage_attach(void) "into secondary processes\n"); } - fd_zero = open("/dev/zero", O_RDONLY); - if (fd_zero < 0) { - RTE_LOG(ERR, EAL, "Could not open /dev/zero\n"); - goto error; - } + test_phys_addrs_available(); + fd_hugepage = open(eal_hugepage_info_path(), O_RDONLY); if (fd_hugepage < 0) { RTE_LOG(ERR, EAL, "Could not open %s\n", eal_hugepage_info_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, MAP_PRIVATE, fd_zero, 0); - if (base_addr == MAP_FAILED || - base_addr != mcfg->memseg[s].addr) { - RTE_LOG(ERR, EAL, "Could not mmap %llu bytes " - "in /dev/zero to requested address [%p]\n", - (unsigned long long)mcfg->memseg[s].len, - mcfg->memseg[s].addr); - 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 == NULL) { + if (hp == MAP_FAILED) { RTE_LOG(ERR, EAL, "Could not mmap %s\n", eal_hugepage_info_path()); goto error; } - num_hp = size / sizeof(struct hugepage); - RTE_LOG(DEBUG, EAL, "Analysing %u hugepages\n", num_hp); + 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; + /* 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; + struct flock lck; + + /* if size is zero, no more pages left */ + if (map_sz == 0) + break; - /* - * 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; - } - addr = mmap(RTE_PTR_ADD(base_addr, offset), - hp[i].size, PROT_READ | PROT_WRITE, - MAP_SHARED | MAP_FIXED, fd, 0); - close(fd); /* close file both on success and on failure */ - if (addr == MAP_FAILED) { - RTE_LOG(ERR, EAL, "Could not mmap %s\n", - hp[i].filepath); - goto error; - } - offset+=hp[i].size; - } + 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)); + goto error; + } + + /* set shared lock on the file. */ + lck.l_type = F_RDLCK; + lck.l_whence = SEEK_SET; + lck.l_start = 0; + lck.l_len = map_sz; + if (fcntl(fd, F_SETLK, &lck) == -1) { + 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((void *)(uintptr_t)hp, size); - close(fd_zero); + munmap(hp, size); close(fd_hugepage); return 0; error: - if (fd_zero >= 0) - close(fd_zero); + /* 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_hugepage >= 0) close(fd_hugepage); return -1; } -static int -rte_eal_memdevice_init(void) +int +rte_eal_hugepage_init(void) { - struct rte_config *config; - - if (rte_eal_process_type() == RTE_PROC_SECONDARY) - return 0; - - config = rte_eal_get_configuration(); - config->mem_config->nchannel = internal_config.force_nchannel; - config->mem_config->nrank = internal_config.force_nrank; - - return 0; + if (internal_config.legacy_mem) + return eal_legacy_hugepage_init(); + return -1; } - -/* init memory subsystem */ int -rte_eal_memory_init(void) +rte_eal_hugepage_attach(void) { - RTE_LOG(INFO, EAL, "Setting up memory...\n"); - const int retval = rte_eal_process_type() == RTE_PROC_PRIMARY ? - rte_eal_hugepage_init() : - rte_eal_hugepage_attach(); - if (retval < 0) - return -1; - - if (internal_config.no_shconf == 0 && rte_eal_memdevice_init() < 0) - return -1; + if (internal_config.legacy_mem) + return eal_legacy_hugepage_attach(); + return -1; +} - return 0; +int +rte_eal_using_phys_addrs(void) +{ + return phys_addrs_available; }