X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=lib%2Flibrte_eal%2Flinuxapp%2Feal%2Feal_memalloc.c;h=79443c56a874048e5bea9a70f59573fbbec63d7e;hb=a23bc2c4e01bf29ed6dd571dbc5b76cc69c10b51;hp=f48d10dcac92ffe1a15e2d43efe784a1742d6a93;hpb=6f0fa9f238320669b024e8f36184b621d9ab2fb3;p=dpdk.git

diff --git a/lib/librte_eal/linuxapp/eal/eal_memalloc.c b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
index f48d10dcac..79443c56a8 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memalloc.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
@@ -28,6 +28,7 @@
 #include <numaif.h>
 #endif
 #include <linux/falloc.h>
+#include <linux/mman.h> /* for hugetlb-related mmap flags */
 
 #include <rte_common.h>
 #include <rte_log.h>
@@ -39,6 +40,16 @@
 #include "eal_filesystem.h"
 #include "eal_internal_cfg.h"
 #include "eal_memalloc.h"
+#include "eal_private.h"
+
+const int anonymous_hugepages_supported =
+#ifdef MAP_HUGE_SHIFT
+		1;
+#define RTE_MAP_HUGE_SHIFT MAP_HUGE_SHIFT
+#else
+		0;
+#define RTE_MAP_HUGE_SHIFT 26
+#endif
 
 /*
  * not all kernel version support fallocate on hugetlbfs, so fall back to
@@ -46,24 +57,25 @@
  */
 static int fallocate_supported = -1; /* unknown */
 
-/*
- * If each page is in a separate file, we can close fd's since we need each fd
- * only once. However, in single file segments mode, we can get away with using
- * a single fd for entire segments, but we need to store them somewhere. Each
- * fd is different within each process, so we'll store them in a local tailq.
+/* for single-file segments, we need some kind of mechanism to keep track of
+ * which hugepages can be freed back to the system, and which cannot. we cannot
+ * use flock() because they don't allow locking parts of a file, and we cannot
+ * use fcntl() due to issues with their semantics, so we will have to rely on a
+ * bunch of lockfiles for each page.
+ *
+ * we cannot know how many pages a system will have in advance, but we do know
+ * that they come in lists, and we know lengths of these lists. so, simply store
+ * a malloc'd array of fd's indexed by list and segment index.
+ *
+ * they will be initialized at startup, and filled as we allocate/deallocate
+ * segments. also, use this to track memseg list proper fd.
  */
-struct msl_entry {
-	TAILQ_ENTRY(msl_entry) next;
-	unsigned int msl_idx;
-	int fd;
-};
-
-/** Double linked list of memseg list fd's. */
-TAILQ_HEAD(msl_entry_list, msl_entry);
-
-static struct msl_entry_list msl_entry_list =
-		TAILQ_HEAD_INITIALIZER(msl_entry_list);
-static rte_spinlock_t tailq_lock = RTE_SPINLOCK_INITIALIZER;
+static struct {
+	int *fds; /**< dynamically allocated array of segment lock fd's */
+	int memseg_list_fd; /**< memseg list fd */
+	int len; /**< total length of the array */
+	int count; /**< entries used in an array */
+} lock_fds[RTE_MAX_MEMSEG_LISTS];
 
 /** local copy of a memory map, used to synchronize memory hotplug in MP */
 static struct rte_memseg_list local_memsegs[RTE_MAX_MEMSEG_LISTS];
@@ -142,11 +154,11 @@ prepare_numa(int *oldpolicy, struct bitmask *oldmask, int socket_id)
 }
 
 static void
-resotre_numa(int *oldpolicy, struct bitmask *oldmask)
+restore_numa(int *oldpolicy, struct bitmask *oldmask)
 {
 	RTE_LOG(DEBUG, EAL,
 		"Restoring previous memory policy: %d\n", *oldpolicy);
-	if (oldpolicy == MPOL_DEFAULT) {
+	if (*oldpolicy == MPOL_DEFAULT) {
 		numa_set_localalloc();
 	} else if (set_mempolicy(*oldpolicy, oldmask->maskp,
 				 oldmask->size + 1) < 0) {
@@ -158,35 +170,6 @@ resotre_numa(int *oldpolicy, struct bitmask *oldmask)
 }
 #endif
 
-static struct msl_entry *
-get_msl_entry_by_idx(unsigned int list_idx)
-{
-	struct msl_entry *te;
-
-	rte_spinlock_lock(&tailq_lock);
-
-	TAILQ_FOREACH(te, &msl_entry_list, next) {
-		if (te->msl_idx == list_idx)
-			break;
-	}
-	if (te == NULL) {
-		/* doesn't exist, so create it and set fd to -1 */
-
-		te = malloc(sizeof(*te));
-		if (te == NULL) {
-			RTE_LOG(ERR, EAL, "%s(): cannot allocate tailq entry for memseg list\n",
-				__func__);
-			goto unlock;
-		}
-		te->msl_idx = list_idx;
-		te->fd = -1;
-		TAILQ_INSERT_TAIL(&msl_entry_list, te, next);
-	}
-unlock:
-	rte_spinlock_unlock(&tailq_lock);
-	return te;
-}
-
 /*
  * uses fstat to report the size of a file on disk
  */
@@ -199,42 +182,99 @@ get_file_size(int fd)
 	return st.st_size;
 }
 
-/*
- * uses fstat to check if file size on disk is zero (regular fstat won't show
- * true file size due to how fallocate works)
- */
-static bool
-is_zero_length(int fd)
+/* returns 1 on successful lock, 0 on unsuccessful lock, -1 on error */
+static int lock(int fd, int type)
 {
-	struct stat st;
-	if (fstat(fd, &st) < 0)
-		return false;
-	return st.st_blocks == 0;
+	int ret;
+
+	/* flock may be interrupted */
+	do {
+		ret = flock(fd, type | LOCK_NB);
+	} while (ret && errno == EINTR);
+
+	if (ret && errno == EWOULDBLOCK) {
+		/* couldn't lock */
+		return 0;
+	} else if (ret) {
+		RTE_LOG(ERR, EAL, "%s(): error calling flock(): %s\n",
+			__func__, strerror(errno));
+		return -1;
+	}
+	/* lock was successful */
+	return 1;
 }
 
-/* we cannot use rte_memseg_list_walk() here because we will be holding a
- * write lock whenever we enter every function in this file, however copying
- * the same iteration code everywhere is not ideal as well. so, use a lockless
- * copy of memseg list walk here.
- */
-static int
-memseg_list_walk_thread_unsafe(rte_memseg_list_walk_t func, void *arg)
+static int get_segment_lock_fd(int list_idx, int seg_idx)
 {
-	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
-	int i, ret = 0;
+	char path[PATH_MAX] = {0};
+	int fd;
 
-	for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
-		struct rte_memseg_list *msl = &mcfg->memsegs[i];
+	if (list_idx < 0 || list_idx >= (int)RTE_DIM(lock_fds))
+		return -1;
+	if (seg_idx < 0 || seg_idx >= lock_fds[list_idx].len)
+		return -1;
 
-		if (msl->base_va == NULL)
-			continue;
+	fd = lock_fds[list_idx].fds[seg_idx];
+	/* does this lock already exist? */
+	if (fd >= 0)
+		return fd;
 
-		ret = func(msl, arg);
-		if (ret < 0)
-			return -1;
-		if (ret > 0)
-			return 1;
+	eal_get_hugefile_lock_path(path, sizeof(path),
+			list_idx * RTE_MAX_MEMSEG_PER_LIST + seg_idx);
+
+	fd = open(path, O_CREAT | O_RDWR, 0660);
+	if (fd < 0) {
+		RTE_LOG(ERR, EAL, "%s(): error creating lockfile '%s': %s\n",
+			__func__, path, strerror(errno));
+		return -1;
+	}
+	/* take out a read lock */
+	if (lock(fd, LOCK_SH) != 1) {
+		RTE_LOG(ERR, EAL, "%s(): failed to take out a readlock on '%s': %s\n",
+			__func__, path, strerror(errno));
+		close(fd);
+		return -1;
+	}
+	/* store it for future reference */
+	lock_fds[list_idx].fds[seg_idx] = fd;
+	lock_fds[list_idx].count++;
+	return fd;
+}
+
+static int unlock_segment(int list_idx, int seg_idx)
+{
+	int fd, ret;
+
+	if (list_idx < 0 || list_idx >= (int)RTE_DIM(lock_fds))
+		return -1;
+	if (seg_idx < 0 || seg_idx >= lock_fds[list_idx].len)
+		return -1;
+
+	fd = lock_fds[list_idx].fds[seg_idx];
+
+	/* upgrade lock to exclusive to see if we can remove the lockfile */
+	ret = lock(fd, LOCK_EX);
+	if (ret == 1) {
+		/* we've succeeded in taking exclusive lock, this lockfile may
+		 * be removed.
+		 */
+		char path[PATH_MAX] = {0};
+		eal_get_hugefile_lock_path(path, sizeof(path),
+				list_idx * RTE_MAX_MEMSEG_PER_LIST + seg_idx);
+		if (unlink(path)) {
+			RTE_LOG(ERR, EAL, "%s(): error removing lockfile '%s': %s\n",
+					__func__, path, strerror(errno));
+		}
 	}
+	/* we don't want to leak the fd, so even if we fail to lock, close fd
+	 * and remove it from list anyway.
+	 */
+	close(fd);
+	lock_fds[list_idx].fds[seg_idx] = -1;
+	lock_fds[list_idx].count--;
+
+	if (ret < 0)
+		return -1;
 	return 0;
 }
 
@@ -245,31 +285,29 @@ get_seg_fd(char *path, int buflen, struct hugepage_info *hi,
 	int fd;
 
 	if (internal_config.single_file_segments) {
-		/*
-		 * try to find a tailq entry, for this memseg list, or create
-		 * one if it doesn't exist.
-		 */
-		struct msl_entry *te = get_msl_entry_by_idx(list_idx);
-		if (te == NULL) {
-			RTE_LOG(ERR, EAL, "%s(): cannot allocate tailq entry for memseg list\n",
-				__func__);
-			return -1;
-		} else if (te->fd < 0) {
-			/* create a hugepage file */
-			eal_get_hugefile_path(path, buflen, hi->hugedir,
-					list_idx);
+		/* create a hugepage file path */
+		eal_get_hugefile_path(path, buflen, hi->hugedir, list_idx);
+
+		fd = lock_fds[list_idx].memseg_list_fd;
+
+		if (fd < 0) {
 			fd = open(path, O_CREAT | O_RDWR, 0600);
 			if (fd < 0) {
-				RTE_LOG(DEBUG, EAL, "%s(): open failed: %s\n",
+				RTE_LOG(ERR, EAL, "%s(): open failed: %s\n",
 					__func__, strerror(errno));
 				return -1;
 			}
-			te->fd = fd;
-		} else {
-			fd = te->fd;
+			/* take out a read lock and keep it indefinitely */
+			if (lock(fd, LOCK_SH) < 0) {
+				RTE_LOG(ERR, EAL, "%s(): lock failed: %s\n",
+					__func__, strerror(errno));
+				close(fd);
+				return -1;
+			}
+			lock_fds[list_idx].memseg_list_fd = fd;
 		}
 	} else {
-		/* one file per page, just create it */
+		/* create a hugepage file path */
 		eal_get_hugefile_path(path, buflen, hi->hugedir,
 				list_idx * RTE_MAX_MEMSEG_PER_LIST + seg_idx);
 		fd = open(path, O_CREAT | O_RDWR, 0600);
@@ -278,48 +316,27 @@ get_seg_fd(char *path, int buflen, struct hugepage_info *hi,
 					strerror(errno));
 			return -1;
 		}
+		/* take out a read lock */
+		if (lock(fd, LOCK_SH) < 0) {
+			RTE_LOG(ERR, EAL, "%s(): lock failed: %s\n",
+				__func__, strerror(errno));
+			close(fd);
+			return -1;
+		}
 	}
 	return fd;
 }
 
-/* returns 1 on successful lock, 0 on unsuccessful lock, -1 on error */
-static int lock(int fd, uint64_t offset, uint64_t len, int type)
-{
-	struct flock lck;
-	int ret;
-
-	memset(&lck, 0, sizeof(lck));
-
-	lck.l_type = type;
-	lck.l_whence = SEEK_SET;
-	lck.l_start = offset;
-	lck.l_len = len;
-
-	ret = fcntl(fd, F_SETLK, &lck);
-
-	if (ret && (errno == EAGAIN || errno == EACCES)) {
-		/* locked by another process, not an error */
-		return 0;
-	} else if (ret) {
-		RTE_LOG(ERR, EAL, "%s(): error calling fcntl(): %s\n",
-			__func__, strerror(errno));
-		/* we've encountered an unexpected error */
-		return -1;
-	}
-	return 1;
-}
-
 static int
-resize_hugefile(int fd, uint64_t fa_offset, uint64_t page_sz,
-		bool grow)
+resize_hugefile(int fd, char *path, int list_idx, int seg_idx,
+		uint64_t fa_offset, uint64_t page_sz, bool grow)
 {
 	bool again = false;
 	do {
 		if (fallocate_supported == 0) {
 			/* we cannot deallocate memory if fallocate() is not
-			 * supported, but locks are still needed to prevent
-			 * primary process' initialization from clearing out
-			 * huge pages used by this process.
+			 * supported, and hugepage file is already locked at
+			 * creation, so no further synchronization needed.
 			 */
 
 			if (!grow) {
@@ -337,13 +354,10 @@ resize_hugefile(int fd, uint64_t fa_offset, uint64_t page_sz,
 					__func__, strerror(errno));
 				return -1;
 			}
-			/* not being able to take out a read lock is an error */
-			if (lock(fd, fa_offset, page_sz, F_RDLCK) != 1)
-				return -1;
 		} else {
 			int flags = grow ? 0 : FALLOC_FL_PUNCH_HOLE |
 					FALLOC_FL_KEEP_SIZE;
-			int ret;
+			int ret, lock_fd;
 
 			/* if fallocate() is supported, we need to take out a
 			 * read lock on allocate (to prevent other processes
@@ -351,20 +365,69 @@ resize_hugefile(int fd, uint64_t fa_offset, uint64_t page_sz,
 			 * lock on deallocate (to ensure nobody else is using
 			 * this page).
 			 *
-			 * we can't use flock() for this, as we actually need to
-			 * lock part of the file, not the entire file.
+			 * read locks on page itself are already taken out at
+			 * file creation, in get_seg_fd().
+			 *
+			 * we cannot rely on simple use of flock() call, because
+			 * we need to be able to lock a section of the file,
+			 * and we cannot use fcntl() locks, because of numerous
+			 * problems with their semantics, so we will use
+			 * deterministically named lock files for each section
+			 * of the file.
+			 *
+			 * if we're shrinking the file, we want to upgrade our
+			 * lock from shared to exclusive.
+			 *
+			 * lock_fd is an fd for a lockfile, not for the segment
+			 * list.
 			 */
+			lock_fd = get_segment_lock_fd(list_idx, seg_idx);
 
 			if (!grow) {
-				ret = lock(fd, fa_offset, page_sz, F_WRLCK);
+				/* we are using this lockfile to determine
+				 * whether this particular page is locked, as we
+				 * are in single file segments mode and thus
+				 * cannot use regular flock() to get this info.
+				 *
+				 * we want to try and take out an exclusive lock
+				 * on the lock file to determine if we're the
+				 * last ones using this page, and if not, we
+				 * won't be shrinking it, and will instead exit
+				 * prematurely.
+				 */
+				ret = lock(lock_fd, LOCK_EX);
+
+				/* drop the lock on the lockfile, so that even
+				 * if we couldn't shrink the file ourselves, we
+				 * are signalling to other processes that we're
+				 * no longer using this page.
+				 */
+				if (unlock_segment(list_idx, seg_idx))
+					RTE_LOG(ERR, EAL, "Could not unlock segment\n");
+
+				/* additionally, if this was the last lock on
+				 * this segment list, we can safely close the
+				 * page file fd, so that one of the processes
+				 * could then delete the file after shrinking.
+				 */
+				if (ret < 1 && lock_fds[list_idx].count == 0) {
+					close(fd);
+					lock_fds[list_idx].memseg_list_fd = -1;
+				}
 
-				if (ret < 0)
+				if (ret < 0) {
+					RTE_LOG(ERR, EAL, "Could not lock segment\n");
 					return -1;
-				else if (ret == 0)
-					/* failed to lock, not an error */
+				}
+				if (ret == 0)
+					/* failed to lock, not an error. */
 					return 0;
 			}
-			if (fallocate(fd, flags, fa_offset, page_sz) < 0) {
+
+			/* grow or shrink the file */
+			ret = fallocate(fd, flags, fa_offset, page_sz);
+
+			if (ret < 0) {
 				if (fallocate_supported == -1 &&
 						errno == ENOTSUP) {
 					RTE_LOG(ERR, EAL, "%s(): fallocate() not supported, hugepage deallocation will be disabled\n",
@@ -380,16 +443,18 @@ resize_hugefile(int fd, uint64_t fa_offset, uint64_t page_sz,
 			} else {
 				fallocate_supported = 1;
 
-				if (grow) {
-					/* if can't read lock, it's an error */
-					if (lock(fd, fa_offset, page_sz,
-							F_RDLCK) != 1)
-						return -1;
-				} else {
-					/* if can't unlock, it's an error */
-					if (lock(fd, fa_offset, page_sz,
-							F_UNLCK) != 1)
-						return -1;
+				/* we've grew/shrunk the file, and we hold an
+				 * exclusive lock now. check if there are no
+				 * more segments active in this segment list,
+				 * and remove the file if there aren't.
+				 */
+				if (lock_fds[list_idx].count == 0) {
+					if (unlink(path))
+						RTE_LOG(ERR, EAL, "%s(): unlinking '%s' failed: %s\n",
+							__func__, path,
+							strerror(errno));
+					close(fd);
+					lock_fds[list_idx].memseg_list_fd = -1;
 				}
 			}
 		}
@@ -406,70 +471,110 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 	int cur_socket_id = 0;
 #endif
 	uint64_t map_offset;
+	rte_iova_t iova;
+	void *va;
 	char path[PATH_MAX];
 	int ret = 0;
 	int fd;
 	size_t alloc_sz;
-
-	fd = get_seg_fd(path, sizeof(path), hi, list_idx, seg_idx);
-	if (fd < 0)
-		return -1;
+	int flags;
+	void *new_addr;
 
 	alloc_sz = hi->hugepage_sz;
-	if (internal_config.single_file_segments) {
-		map_offset = seg_idx * alloc_sz;
-		ret = resize_hugefile(fd, map_offset, alloc_sz, true);
-		if (ret < 0)
-			goto resized;
-	} else {
+	if (internal_config.in_memory && anonymous_hugepages_supported) {
+		int log2, flags;
+
+		log2 = rte_log2_u32(alloc_sz);
+		/* as per mmap() manpage, all page sizes are log2 of page size
+		 * shifted by MAP_HUGE_SHIFT
+		 */
+		flags = (log2 << RTE_MAP_HUGE_SHIFT) | MAP_HUGETLB | MAP_FIXED |
+				MAP_PRIVATE | MAP_ANONYMOUS;
+		fd = -1;
+		va = mmap(addr, alloc_sz, PROT_READ | PROT_WRITE, flags, -1, 0);
+
+		/* single-file segments codepath will never be active because
+		 * in-memory mode is incompatible with it and it's stopped at
+		 * EAL initialization stage, however the compiler doesn't know
+		 * that and complains about map_offset being used uninitialized
+		 * on failure codepaths while having in-memory mode enabled. so,
+		 * assign a value here.
+		 */
 		map_offset = 0;
-		if (ftruncate(fd, alloc_sz) < 0) {
-			RTE_LOG(DEBUG, EAL, "%s(): ftruncate() failed: %s\n",
-				__func__, strerror(errno));
-			goto resized;
+	} else {
+		/* takes out a read lock on segment or segment list */
+		fd = get_seg_fd(path, sizeof(path), hi, list_idx, seg_idx);
+		if (fd < 0) {
+			RTE_LOG(ERR, EAL, "Couldn't get fd on hugepage file\n");
+			return -1;
 		}
-		/* we've allocated a page - take out a read lock. we're using
-		 * fcntl() locks rather than flock() here because doing that
-		 * gives us one huge advantage - fcntl() locks are per-process,
-		 * not per-file descriptor, which means that we don't have to
-		 * keep the original fd's around to keep a lock on the file.
-		 *
-		 * this is useful, because when it comes to unmapping pages, we
-		 * will have to take out a write lock (to figure out if another
-		 * process still has this page mapped), and to do itwith flock()
-		 * we'll have to use original fd, as lock is associated with
-		 * that particular fd. with fcntl(), this is not necessary - we
-		 * can open a new fd and use fcntl() on that.
-		 */
-		ret = lock(fd, map_offset, alloc_sz, F_RDLCK);
-
-		/* this should not fail */
-		if (ret != 1) {
-			RTE_LOG(ERR, EAL, "%s(): error locking file: %s\n",
-				__func__,
-				strerror(errno));
-			goto resized;
+
+		if (internal_config.single_file_segments) {
+			map_offset = seg_idx * alloc_sz;
+			ret = resize_hugefile(fd, path, list_idx, seg_idx,
+					map_offset, alloc_sz, true);
+			if (ret < 0)
+				goto resized;
+		} else {
+			map_offset = 0;
+			if (ftruncate(fd, alloc_sz) < 0) {
+				RTE_LOG(DEBUG, EAL, "%s(): ftruncate() failed: %s\n",
+					__func__, strerror(errno));
+				goto resized;
+			}
+			if (internal_config.hugepage_unlink) {
+				if (unlink(path)) {
+					RTE_LOG(DEBUG, EAL, "%s(): unlink() failed: %s\n",
+						__func__, strerror(errno));
+					goto resized;
+				}
+			}
 		}
-	}
 
-	/*
-	 * map the segment, and populate page tables, the kernel fills this
-	 * segment with zeros if it's a new page.
-	 */
-	void *va = mmap(addr, alloc_sz, PROT_READ | PROT_WRITE,
-			MAP_SHARED | MAP_POPULATE | MAP_FIXED, fd, map_offset);
+		/*
+		 * map the segment, and populate page tables, the kernel fills
+		 * this segment with zeros if it's a new page.
+		 */
+		va = mmap(addr, alloc_sz, PROT_READ | PROT_WRITE,
+				MAP_SHARED | MAP_POPULATE | MAP_FIXED, fd,
+				map_offset);
+	}
 
 	if (va == MAP_FAILED) {
 		RTE_LOG(DEBUG, EAL, "%s(): mmap() failed: %s\n", __func__,
 			strerror(errno));
-		goto resized;
+		/* mmap failed, but the previous region might have been
+		 * unmapped anyway. try to remap it
+		 */
+		goto unmapped;
 	}
 	if (va != addr) {
 		RTE_LOG(DEBUG, EAL, "%s(): wrong mmap() address\n", __func__);
+		munmap(va, alloc_sz);
+		goto resized;
+	}
+
+	/* 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 %uMB\n",
+			(unsigned int)(alloc_sz >> 20));
 		goto mapped;
 	}
 
-	rte_iova_t iova = rte_mem_virt2iova(addr);
+	/* we need to trigger a write to the page to enforce page fault and
+	 * ensure that page is accessible to us, but we can't overwrite value
+	 * that is already there, so read the old value, and write itback.
+	 * kernel populates the page with zeroes initially.
+	 */
+	*(volatile int *)addr = *(volatile int *)addr;
+
+	iova = rte_mem_virt2iova(addr);
 	if (iova == RTE_BAD_PHYS_ADDR) {
 		RTE_LOG(DEBUG, EAL, "%s(): can't get IOVA addr\n",
 			__func__);
@@ -486,25 +591,11 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 		goto mapped;
 	}
 #endif
-
-	/* 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 %uMB\n",
-			(unsigned int)(alloc_sz >> 20));
-		goto mapped;
-	}
-	/* for non-single file segments, we can close fd here */
-	if (!internal_config.single_file_segments)
+	/* for non-single file segments that aren't in-memory, we can close fd
+	 * here */
+	if (!internal_config.single_file_segments && !internal_config.in_memory)
 		close(fd);
 
-	*(int *)addr = *(int *)addr;
-
 	ms->addr = addr;
 	ms->hugepage_sz = alloc_sz;
 	ms->len = alloc_sz;
@@ -517,21 +608,34 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 
 mapped:
 	munmap(addr, alloc_sz);
+unmapped:
+	flags = MAP_FIXED;
+#ifdef RTE_ARCH_PPC_64
+	flags |= MAP_HUGETLB;
+#endif
+	new_addr = eal_get_virtual_area(addr, &alloc_sz, alloc_sz, 0, flags);
+	if (new_addr != addr) {
+		if (new_addr != NULL)
+			munmap(new_addr, alloc_sz);
+		/* we're leaving a hole in our virtual address space. if
+		 * somebody else maps this hole now, we could accidentally
+		 * override it in the future.
+		 */
+		RTE_LOG(CRIT, EAL, "Can't mmap holes in our virtual address space\n");
+	}
 resized:
+	/* in-memory mode will never be single-file-segments mode */
 	if (internal_config.single_file_segments) {
-		resize_hugefile(fd, map_offset, alloc_sz, false);
-		if (is_zero_length(fd)) {
-			struct msl_entry *te = get_msl_entry_by_idx(list_idx);
-			if (te != NULL && te->fd >= 0) {
-				close(te->fd);
-				te->fd = -1;
-			}
-			/* ignore errors, can't make it any worse */
-			unlink(path);
-		}
+		resize_hugefile(fd, path, list_idx, seg_idx, map_offset,
+				alloc_sz, false);
+		/* ignore failure, can't make it any worse */
 	} else {
+		/* only remove file if we can take out a write lock */
+		if (internal_config.hugepage_unlink == 0 &&
+				internal_config.in_memory == 0 &&
+				lock(fd, LOCK_EX) == 1)
+			unlink(path);
 		close(fd);
-		unlink(path);
 	}
 	return -1;
 }
@@ -554,47 +658,43 @@ free_seg(struct rte_memseg *ms, struct hugepage_info *hi,
 		return -1;
 	}
 
+	/* if we've already unlinked the page, nothing needs to be done */
+	if (internal_config.hugepage_unlink) {
+		memset(ms, 0, sizeof(*ms));
+		return 0;
+	}
+
+	/* if we are not in single file segments mode, we're going to unmap the
+	 * segment and thus drop the lock on original fd, but hugepage dir is
+	 * now locked so we can take out another one without races.
+	 */
 	fd = get_seg_fd(path, sizeof(path), hi, list_idx, seg_idx);
 	if (fd < 0)
 		return -1;
 
 	if (internal_config.single_file_segments) {
 		map_offset = seg_idx * ms->len;
-		if (resize_hugefile(fd, map_offset, ms->len, false))
+		if (resize_hugefile(fd, path, list_idx, seg_idx, map_offset,
+				ms->len, false))
 			return -1;
-		/* if file is zero-length, we've already shrunk it, so it's
-		 * safe to remove.
-		 */
-		if (is_zero_length(fd)) {
-			struct msl_entry *te = get_msl_entry_by_idx(list_idx);
-			if (te != NULL && te->fd >= 0) {
-				close(te->fd);
-				te->fd = -1;
-			}
-			unlink(path);
-		}
 		ret = 0;
 	} else {
 		/* if we're able to take out a write lock, we're the last one
 		 * holding onto this page.
 		 */
-
-		ret = lock(fd, 0, ms->len, F_WRLCK);
+		ret = lock(fd, LOCK_EX);
 		if (ret >= 0) {
 			/* no one else is using this page */
 			if (ret == 1)
 				unlink(path);
-			ret = lock(fd, 0, ms->len, F_UNLCK);
-			if (ret != 1)
-				RTE_LOG(ERR, EAL, "%s(): unable to unlock file %s\n",
-					__func__, path);
 		}
+		/* closing fd will drop the lock */
 		close(fd);
 	}
 
 	memset(ms, 0, sizeof(*ms));
 
-	return ret;
+	return ret < 0 ? -1 : 0;
 }
 
 struct alloc_walk_param {
@@ -613,7 +713,7 @@ alloc_seg_walk(const struct rte_memseg_list *msl, void *arg)
 	struct alloc_walk_param *wa = arg;
 	struct rte_memseg_list *cur_msl;
 	size_t page_sz;
-	int cur_idx, start_idx, j;
+	int cur_idx, start_idx, j, dir_fd = -1;
 	unsigned int msl_idx, need, i;
 
 	if (msl->page_sz != wa->page_sz)
@@ -629,11 +729,36 @@ alloc_seg_walk(const struct rte_memseg_list *msl, void *arg)
 	need = wa->n_segs;
 
 	/* try finding space in memseg list */
-	cur_idx = rte_fbarray_find_next_n_free(&cur_msl->memseg_arr, 0, need);
+	cur_idx = rte_fbarray_find_prev_n_free(&cur_msl->memseg_arr,
+			cur_msl->memseg_arr.len - 1, need);
 	if (cur_idx < 0)
 		return 0;
 	start_idx = cur_idx;
 
+	/* do not allow any page allocations during the time we're allocating,
+	 * because file creation and locking operations are not atomic,
+	 * and we might be the first or the last ones to use a particular page,
+	 * so we need to ensure atomicity of every operation.
+	 *
+	 * during init, we already hold a write lock, so don't try to take out
+	 * another one.
+	 */
+	if (wa->hi->lock_descriptor == -1 && !internal_config.in_memory) {
+		dir_fd = open(wa->hi->hugedir, O_RDONLY);
+		if (dir_fd < 0) {
+			RTE_LOG(ERR, EAL, "%s(): Cannot open '%s': %s\n",
+				__func__, wa->hi->hugedir, strerror(errno));
+			return -1;
+		}
+		/* blocking writelock */
+		if (flock(dir_fd, LOCK_EX)) {
+			RTE_LOG(ERR, EAL, "%s(): Cannot lock '%s': %s\n",
+				__func__, wa->hi->hugedir, strerror(errno));
+			close(dir_fd);
+			return -1;
+		}
+	}
+
 	for (i = 0; i < need; i++, cur_idx++) {
 		struct rte_memseg *cur;
 		void *map_addr;
@@ -658,17 +783,20 @@ alloc_seg_walk(const struct rte_memseg_list *msl, void *arg)
 						&cur_msl->memseg_arr;
 
 				tmp = rte_fbarray_get(arr, j);
-				if (free_seg(tmp, wa->hi, msl_idx,
-						start_idx + j)) {
-					RTE_LOG(ERR, EAL, "Cannot free page\n");
-					continue;
-				}
-
 				rte_fbarray_set_free(arr, j);
+
+				/* free_seg may attempt to create a file, which
+				 * may fail.
+				 */
+				if (free_seg(tmp, wa->hi, msl_idx, j))
+					RTE_LOG(DEBUG, EAL, "Cannot free page\n");
 			}
 			/* clear the list */
 			if (wa->ms)
 				memset(wa->ms, 0, sizeof(*wa->ms) * wa->n_segs);
+
+			if (dir_fd >= 0)
+				close(dir_fd);
 			return -1;
 		}
 		if (wa->ms)
@@ -680,6 +808,8 @@ out:
 	wa->segs_allocated = i;
 	if (i > 0)
 		cur_msl->version++;
+	if (dir_fd >= 0)
+		close(dir_fd);
 	return 1;
 }
 
@@ -694,7 +824,7 @@ free_seg_walk(const struct rte_memseg_list *msl, void *arg)
 	struct rte_memseg_list *found_msl;
 	struct free_walk_param *wa = arg;
 	uintptr_t start_addr, end_addr;
-	int msl_idx, seg_idx;
+	int msl_idx, seg_idx, ret, dir_fd = -1;
 
 	start_addr = (uintptr_t) msl->base_va;
 	end_addr = start_addr + msl->memseg_arr.len * (size_t)msl->page_sz;
@@ -709,11 +839,40 @@ free_seg_walk(const struct rte_memseg_list *msl, void *arg)
 	/* msl is const */
 	found_msl = &mcfg->memsegs[msl_idx];
 
+	/* do not allow any page allocations during the time we're freeing,
+	 * because file creation and locking operations are not atomic,
+	 * and we might be the first or the last ones to use a particular page,
+	 * so we need to ensure atomicity of every operation.
+	 *
+	 * during init, we already hold a write lock, so don't try to take out
+	 * another one.
+	 */
+	if (wa->hi->lock_descriptor == -1 && !internal_config.in_memory) {
+		dir_fd = open(wa->hi->hugedir, O_RDONLY);
+		if (dir_fd < 0) {
+			RTE_LOG(ERR, EAL, "%s(): Cannot open '%s': %s\n",
+				__func__, wa->hi->hugedir, strerror(errno));
+			return -1;
+		}
+		/* blocking writelock */
+		if (flock(dir_fd, LOCK_EX)) {
+			RTE_LOG(ERR, EAL, "%s(): Cannot lock '%s': %s\n",
+				__func__, wa->hi->hugedir, strerror(errno));
+			close(dir_fd);
+			return -1;
+		}
+	}
+
 	found_msl->version++;
 
 	rte_fbarray_set_free(&found_msl->memseg_arr, seg_idx);
 
-	if (free_seg(wa->ms, wa->hi, msl_idx, seg_idx))
+	ret = free_seg(wa->ms, wa->hi, msl_idx, seg_idx);
+
+	if (dir_fd >= 0)
+		close(dir_fd);
+
+	if (ret < 0)
 		return -1;
 
 	return 1;
@@ -767,7 +926,8 @@ eal_memalloc_alloc_seg_bulk(struct rte_memseg **ms, int n_segs, size_t page_sz,
 	wa.socket = socket;
 	wa.segs_allocated = 0;
 
-	ret = memseg_list_walk_thread_unsafe(alloc_seg_walk, &wa);
+	/* memalloc is locked, so it's safe to use thread-unsafe version */
+	ret = rte_memseg_list_walk_thread_unsafe(alloc_seg_walk, &wa);
 	if (ret == 0) {
 		RTE_LOG(ERR, EAL, "%s(): couldn't find suitable memseg_list\n",
 			__func__);
@@ -778,7 +938,7 @@ eal_memalloc_alloc_seg_bulk(struct rte_memseg **ms, int n_segs, size_t page_sz,
 
 #ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES
 	if (have_numa)
-		resotre_numa(&oldpolicy, oldmask);
+		restore_numa(&oldpolicy, oldmask);
 #endif
 	return ret;
 }
@@ -832,7 +992,10 @@ eal_memalloc_free_seg_bulk(struct rte_memseg **ms, int n_segs)
 		wa.ms = cur;
 		wa.hi = hi;
 
-		walk_res = memseg_list_walk_thread_unsafe(free_seg_walk, &wa);
+		/* memalloc is locked, so it's safe to use thread-unsafe version
+		 */
+		walk_res = rte_memseg_list_walk_thread_unsafe(free_seg_walk,
+				&wa);
 		if (walk_res == 1)
 			continue;
 		if (walk_res == 0)
@@ -1035,22 +1198,46 @@ sync_existing(struct rte_memseg_list *primary_msl,
 		struct rte_memseg_list *local_msl, struct hugepage_info *hi,
 		unsigned int msl_idx)
 {
-	int ret;
+	int ret, dir_fd;
+
+	/* do not allow any page allocations during the time we're allocating,
+	 * because file creation and locking operations are not atomic,
+	 * and we might be the first or the last ones to use a particular page,
+	 * so we need to ensure atomicity of every operation.
+	 */
+	dir_fd = open(hi->hugedir, O_RDONLY);
+	if (dir_fd < 0) {
+		RTE_LOG(ERR, EAL, "%s(): Cannot open '%s': %s\n", __func__,
+			hi->hugedir, strerror(errno));
+		return -1;
+	}
+	/* blocking writelock */
+	if (flock(dir_fd, LOCK_EX)) {
+		RTE_LOG(ERR, EAL, "%s(): Cannot lock '%s': %s\n", __func__,
+			hi->hugedir, strerror(errno));
+		close(dir_fd);
+		return -1;
+	}
 
 	/* ensure all allocated space is the same in both lists */
 	ret = sync_status(primary_msl, local_msl, hi, msl_idx, true);
 	if (ret < 0)
-		return -1;
+		goto fail;
 
 	/* ensure all unallocated space is the same in both lists */
 	ret = sync_status(primary_msl, local_msl, hi, msl_idx, false);
 	if (ret < 0)
-		return -1;
+		goto fail;
 
 	/* update version number */
 	local_msl->version = primary_msl->version;
 
+	close(dir_fd);
+
 	return 0;
+fail:
+	close(dir_fd);
+	return -1;
 }
 
 static int
@@ -1061,33 +1248,11 @@ sync_walk(const struct rte_memseg_list *msl, void *arg __rte_unused)
 	struct hugepage_info *hi = NULL;
 	unsigned int i;
 	int msl_idx;
-	bool new_msl = false;
 
 	msl_idx = msl - mcfg->memsegs;
 	primary_msl = &mcfg->memsegs[msl_idx];
 	local_msl = &local_memsegs[msl_idx];
 
-	/* check if secondary has this memseg list set up */
-	if (local_msl->base_va == NULL) {
-		char name[PATH_MAX];
-		int ret;
-		new_msl = true;
-
-		/* create distinct fbarrays for each secondary */
-		snprintf(name, RTE_FBARRAY_NAME_LEN, "%s_%i",
-			primary_msl->memseg_arr.name, getpid());
-
-		ret = rte_fbarray_init(&local_msl->memseg_arr, name,
-			primary_msl->memseg_arr.len,
-			primary_msl->memseg_arr.elt_sz);
-		if (ret < 0) {
-			RTE_LOG(ERR, EAL, "Cannot initialize local memory map\n");
-			return -1;
-		}
-
-		local_msl->base_va = primary_msl->base_va;
-	}
-
 	for (i = 0; i < RTE_DIM(internal_config.hugepage_info); i++) {
 		uint64_t cur_sz =
 			internal_config.hugepage_info[i].hugepage_sz;
@@ -1102,10 +1267,8 @@ sync_walk(const struct rte_memseg_list *msl, void *arg __rte_unused)
 		return -1;
 	}
 
-	/* if versions don't match or if we have just allocated a new
-	 * memseg list, synchronize everything
-	 */
-	if ((new_msl || local_msl->version != primary_msl->version) &&
+	/* if versions don't match, synchronize everything */
+	if (local_msl->version != primary_msl->version &&
 			sync_existing(primary_msl, local_msl, hi, msl_idx))
 		return -1;
 	return 0;
@@ -1119,7 +1282,81 @@ eal_memalloc_sync_with_primary(void)
 	if (rte_eal_process_type() == RTE_PROC_PRIMARY)
 		return 0;
 
-	if (memseg_list_walk_thread_unsafe(sync_walk, NULL))
+	/* memalloc is locked, so it's safe to call thread-unsafe version */
+	if (rte_memseg_list_walk_thread_unsafe(sync_walk, NULL))
 		return -1;
 	return 0;
 }
+
+static int
+secondary_msl_create_walk(const struct rte_memseg_list *msl,
+		void *arg __rte_unused)
+{
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+	struct rte_memseg_list *primary_msl, *local_msl;
+	char name[PATH_MAX];
+	int msl_idx, ret;
+
+	msl_idx = msl - mcfg->memsegs;
+	primary_msl = &mcfg->memsegs[msl_idx];
+	local_msl = &local_memsegs[msl_idx];
+
+	/* create distinct fbarrays for each secondary */
+	snprintf(name, RTE_FBARRAY_NAME_LEN, "%s_%i",
+		primary_msl->memseg_arr.name, getpid());
+
+	ret = rte_fbarray_init(&local_msl->memseg_arr, name,
+		primary_msl->memseg_arr.len,
+		primary_msl->memseg_arr.elt_sz);
+	if (ret < 0) {
+		RTE_LOG(ERR, EAL, "Cannot initialize local memory map\n");
+		return -1;
+	}
+	local_msl->base_va = primary_msl->base_va;
+
+	return 0;
+}
+
+static int
+secondary_lock_list_create_walk(const struct rte_memseg_list *msl,
+		void *arg __rte_unused)
+{
+	struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+	unsigned int i, len;
+	int msl_idx;
+	int *data;
+
+	msl_idx = msl - mcfg->memsegs;
+	len = msl->memseg_arr.len;
+
+	/* ensure we have space to store lock fd per each possible segment */
+	data = malloc(sizeof(int) * len);
+	if (data == NULL) {
+		RTE_LOG(ERR, EAL, "Unable to allocate space for lock descriptors\n");
+		return -1;
+	}
+	/* set all fd's as invalid */
+	for (i = 0; i < len; i++)
+		data[i] = -1;
+
+	lock_fds[msl_idx].fds = data;
+	lock_fds[msl_idx].len = len;
+	lock_fds[msl_idx].count = 0;
+	lock_fds[msl_idx].memseg_list_fd = -1;
+
+	return 0;
+}
+
+int
+eal_memalloc_init(void)
+{
+	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
+		if (rte_memseg_list_walk(secondary_msl_create_walk, NULL) < 0)
+			return -1;
+
+	/* initialize all of the lock fd lists */
+	if (internal_config.single_file_segments)
+		if (rte_memseg_list_walk(secondary_lock_list_create_walk, NULL))
+			return -1;
+	return 0;
+}