[dpdk-dev] [PATCH 8/8] mem: support using memfd segments for in-memory mode

[Anatoly Burakov <anatoly.burakov@intel.com>]

Enable using memfd-created segments if supported by the system.

This will allow having real fd's for pages but without hugetlbfs
mounts, which will enable in-memory mode to be used with virtio.

The implementation is mostly piggy-backing on existing real-fd
code, except that we no longer need to unlink any files or track
per-page locks in single-file segments mode, because in-memory
mode does not support secondary processes anyway.

We move some checks from EAL command-line parsing code to memalloc
because it is now possible to use single-file segments mode with
in-memory mode, but only if memfd is supported.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
 lib/librte_eal/common/eal_common_options.c |   6 +-
 lib/librte_eal/linuxapp/eal/eal_memalloc.c | 265 ++++++++++++++++++---
 2 files changed, 235 insertions(+), 36 deletions(-)

diff --git a/lib/librte_eal/common/eal_common_options.c b/lib/librte_eal/common/eal_common_options.c
index 873099acc..ddd624110 100644
--- a/lib/librte_eal/common/eal_common_options.c
+++ b/lib/librte_eal/common/eal_common_options.c
@@ -1384,10 +1384,10 @@ eal_check_common_options(struct internal_config *internal_cfg)
 			" is only supported in non-legacy memory mode\n");
 	}
 	if (internal_cfg->single_file_segments &&
-			internal_cfg->hugepage_unlink) {
+			internal_cfg->hugepage_unlink &&
+			!internal_cfg->in_memory) {
 		RTE_LOG(ERR, EAL, "Option --"OPT_SINGLE_FILE_SEGMENTS" is "
-			"not compatible with neither --"OPT_IN_MEMORY" nor "
-			"--"OPT_HUGE_UNLINK"\n");
+			"not compatible with --"OPT_HUGE_UNLINK"\n");
 		return -1;
 	}
 	if (internal_cfg->legacy_mem &&
diff --git a/lib/librte_eal/linuxapp/eal/eal_memalloc.c b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
index b820989e9..50379136c 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memalloc.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memalloc.c
@@ -51,6 +51,23 @@ const int anonymous_hugepages_supported =
 #define RTE_MAP_HUGE_SHIFT 26
 #endif
 
+/*
+ * we don't actually care if memfd itself is supported - we only need to check
+ * if memfd supports hugetlbfs, as that already implies memfd support.
+ *
+ * also, this is not a constant, because while we may be *compiled* with memfd
+ * hugetlbfs support, we might not be *running* on a system that supports memfd
+ * and/or memfd with hugetlbfs, so we need to be able to adjust this flag at
+ * runtime, and fall back to anonymous memory.
+ */
+int memfd_create_supported =
+#ifdef MFD_HUGETLB
+#define MEMFD_SUPPORTED
+		1;
+#else
+		0;
+#endif
+
 /*
  * not all kernel version support fallocate on hugetlbfs, so fall back to
  * ftruncate and disallow deallocation if fallocate is not supported.
@@ -190,6 +207,31 @@ get_file_size(int fd)
 	return st.st_size;
 }
 
+static inline uint32_t
+bsf64(uint64_t v)
+{
+	return (uint32_t)__builtin_ctzll(v);
+}
+
+static inline uint32_t
+log2_u64(uint64_t v)
+{
+	if (v == 0)
+		return 0;
+	v = rte_align64pow2(v);
+	return bsf64(v);
+}
+
+static int
+pagesz_flags(uint64_t page_sz)
+{
+	/* as per mmap() manpage, all page sizes are log2 of page size
+	 * shifted by MAP_HUGE_SHIFT
+	 */
+	int log2 = log2_u64(page_sz);
+	return log2 << RTE_MAP_HUGE_SHIFT;
+}
+
 /* returns 1 on successful lock, 0 on unsuccessful lock, -1 on error */
 static int lock(int fd, int type)
 {
@@ -286,12 +328,64 @@ static int unlock_segment(int list_idx, int seg_idx)
 	return 0;
 }
 
+static int
+get_seg_memfd(struct hugepage_info *hi __rte_unused,
+		unsigned int list_idx __rte_unused,
+		unsigned int seg_idx __rte_unused)
+{
+#ifdef MEMFD_SUPPORTED
+	int fd;
+	char segname[250]; /* as per manpage, limit is 249 bytes plus null */
+
+	if (internal_config.single_file_segments) {
+		fd = fd_list[list_idx].memseg_list_fd;
+
+		if (fd < 0) {
+			int flags = MFD_HUGETLB | pagesz_flags(hi->hugepage_sz);
+
+			snprintf(segname, sizeof(segname), "seg_%i", list_idx);
+			fd = memfd_create(segname, flags);
+			if (fd < 0) {
+				RTE_LOG(DEBUG, EAL, "%s(): memfd create failed: %s\n",
+					__func__, strerror(errno));
+				return -1;
+			}
+			fd_list[list_idx].memseg_list_fd = fd;
+		}
+	} else {
+		fd = fd_list[list_idx].fds[seg_idx];
+
+		if (fd < 0) {
+			int flags = MFD_HUGETLB | pagesz_flags(hi->hugepage_sz);
+
+			snprintf(segname, sizeof(segname), "seg_%i-%i",
+					list_idx, seg_idx);
+			fd = memfd_create(segname, flags);
+			if (fd < 0) {
+				RTE_LOG(DEBUG, EAL, "%s(): memfd create failed: %s\n",
+					__func__, strerror(errno));
+				return -1;
+			}
+			fd_list[list_idx].fds[seg_idx] = fd;
+		}
+	}
+	return fd;
+#endif
+	return -1;
+}
+
 static int
 get_seg_fd(char *path, int buflen, struct hugepage_info *hi,
 		unsigned int list_idx, unsigned int seg_idx)
 {
 	int fd;
 
+	/* for in-memory mode, we only make it here when we're sure we support
+	 * memfd, and this is a special case.
+	 */
+	if (internal_config.in_memory)
+		return get_seg_memfd(hi, list_idx, seg_idx);
+
 	if (internal_config.single_file_segments) {
 		/* create a hugepage file path */
 		eal_get_hugefile_path(path, buflen, hi->hugedir, list_idx);
@@ -346,6 +440,33 @@ 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;
+
+	/* in-memory mode is a special case, because we don't need to perform
+	 * any locking, and we can be sure that fallocate() is supported.
+	 */
+	if (internal_config.in_memory) {
+		int flags = grow ? 0 : FALLOC_FL_PUNCH_HOLE |
+				FALLOC_FL_KEEP_SIZE;
+		int ret;
+
+		/* grow or shrink the file */
+		ret = fallocate(fd, flags, fa_offset, page_sz);
+
+		if (ret < 0) {
+			RTE_LOG(DEBUG, EAL, "%s(): fallocate() failed: %s\n",
+					__func__,
+					strerror(errno));
+			return -1;
+		}
+		/* increase/decrease total segment count */
+		fd_list[list_idx].count += (grow ? 1 : -1);
+		if (!grow && fd_list[list_idx].count == 0) {
+			close(fd_list[list_idx].memseg_list_fd);
+			fd_list[list_idx].memseg_list_fd = -1;
+		}
+		return 0;
+	}
+
 	do {
 		if (fallocate_supported == 0) {
 			/* we cannot deallocate memory if fallocate() is not
@@ -495,26 +616,34 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 	void *new_addr;
 
 	alloc_sz = hi->hugepage_sz;
-	if (!internal_config.single_file_segments &&
-			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 |
+
+	/* these are checked at init, but code analyzers don't know that */
+	if (internal_config.in_memory && !anonymous_hugepages_supported) {
+		RTE_LOG(ERR, EAL, "Anonymous hugepages not supported, in-memory mode cannot allocate memory\n");
+		return -1;
+	}
+	if (internal_config.in_memory && !memfd_create_supported &&
+			internal_config.single_file_segments) {
+		RTE_LOG(ERR, EAL, "Single-file segments are not supported without memfd support\n");
+		return -1;
+	}
+
+	/* in-memory without memfd is a special case */
+	int mmap_flags;
+
+	if (internal_config.in_memory && !memfd_create_supported) {
+		int pagesz_flag, flags;
+
+		pagesz_flag = pagesz_flags(alloc_sz);
+		flags = pagesz_flag | MAP_HUGETLB | MAP_FIXED |
 				MAP_PRIVATE | MAP_ANONYMOUS;
 		fd = -1;
-		va = mmap(addr, alloc_sz, PROT_READ | PROT_WRITE, flags, -1, 0);
+		mmap_flags = flags;
 
-		/* 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.
+		/* single-file segments codepath will never be active
+		 * here because in-memory mode is incompatible with the
+		 * fallback path, and it's stopped at EAL initialization
+		 * stage.
 		 */
 		map_offset = 0;
 	} else {
@@ -538,7 +667,8 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 					__func__, strerror(errno));
 				goto resized;
 			}
-			if (internal_config.hugepage_unlink) {
+			if (internal_config.hugepage_unlink &&
+					!internal_config.in_memory) {
 				if (unlink(path)) {
 					RTE_LOG(DEBUG, EAL, "%s(): unlink() failed: %s\n",
 						__func__, strerror(errno));
@@ -546,16 +676,16 @@ alloc_seg(struct rte_memseg *ms, void *addr, int socket_id,
 				}
 			}
 		}
-
-		/*
-		 * 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);
+		mmap_flags = MAP_SHARED | MAP_POPULATE | MAP_FIXED;
 	}
 
+	/*
+	 * 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, mmap_flags, fd,
+			map_offset);
+
 	if (va == MAP_FAILED) {
 		RTE_LOG(DEBUG, EAL, "%s(): mmap() failed: %s\n", __func__,
 			strerror(errno));
@@ -662,7 +792,8 @@ free_seg(struct rte_memseg *ms, struct hugepage_info *hi,
 {
 	uint64_t map_offset;
 	char path[PATH_MAX];
-	int fd, ret;
+	int fd, ret = 0;
+	bool exit_early;
 
 	/* erase page data */
 	memset(ms->addr, 0, ms->len);
@@ -674,8 +805,17 @@ free_seg(struct rte_memseg *ms, struct hugepage_info *hi,
 		return -1;
 	}
 
+	exit_early = false;
+
+	/* if we're using anonymous hugepages, nothing to be done */
+	if (internal_config.in_memory && !memfd_create_supported)
+		exit_early = true;
+
 	/* if we've already unlinked the page, nothing needs to be done */
-	if (internal_config.hugepage_unlink) {
+	if (!internal_config.in_memory && internal_config.hugepage_unlink)
+		exit_early = true;
+
+	if (exit_early) {
 		memset(ms, 0, sizeof(*ms));
 		return 0;
 	}
@@ -698,11 +838,13 @@ free_seg(struct rte_memseg *ms, struct hugepage_info *hi,
 		/* if we're able to take out a write lock, we're the last one
 		 * holding onto this page.
 		 */
-		ret = lock(fd, LOCK_EX);
-		if (ret >= 0) {
-			/* no one else is using this page */
-			if (ret == 1)
-				unlink(path);
+		if (!internal_config.in_memory) {
+			ret = lock(fd, LOCK_EX);
+			if (ret >= 0) {
+				/* no one else is using this page */
+				if (ret == 1)
+					unlink(path);
+			}
 		}
 		/* closing fd will drop the lock */
 		close(fd);
@@ -1399,12 +1541,69 @@ eal_memalloc_get_seg_fd(int list_idx, int seg_idx)
 	return fd_list[list_idx].fds[seg_idx];
 }
 
+static int
+test_memfd_create(void)
+{
+#ifdef MEMFD_SUPPORTED
+	unsigned int i;
+	for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
+		uint64_t pagesz = internal_config.hugepage_info[i].hugepage_sz;
+		int pagesz_flag = pagesz_flags(pagesz);
+		int flags;
+
+		flags = pagesz_flag | MFD_HUGETLB;
+		int fd = memfd_create("test", flags);
+		if (fd < 0) {
+			/* we failed - let memalloc know this isn't working */
+			if (errno == EINVAL) {
+				memfd_create_supported = 0;
+				return 0; /* not supported */
+			}
+
+			/* we got other error - something's wrong */
+			return -1; /* error */
+		}
+		close(fd);
+		return 1; /* supported */
+	}
+#endif
+	return 0; /* not supported */
+}
+
 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;
+	if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
+			internal_config.in_memory) {
+		int mfd_res = test_memfd_create();
+
+		if (mfd_res < 0) {
+			RTE_LOG(ERR, EAL, "Unable to check if memfd is supported\n");
+			return -1;
+		}
+		if (mfd_res == 1)
+			RTE_LOG(DEBUG, EAL, "Using memfd for anonymous memory\n");
+		else
+			RTE_LOG(INFO, EAL, "Using memfd is not supported, falling back to anonymous hugepages\n");
+
+		/* we only support single-file segments mode with in-memory mode
+		 * if we support hugetlbfs with memfd_create. this code will
+		 * test if we do.
+		 */
+		if (internal_config.single_file_segments &&
+				mfd_res != 1) {
+			RTE_LOG(ERR, EAL, "Single-file segments mode cannot be used without memfd support\n");
+			return -1;
+		}
+		/* this cannot ever happen but better safe than sorry */
+		if (!anonymous_hugepages_supported) {
+			RTE_LOG(ERR, EAL, "Using anonymous memory is not supported\n");
+			return -1;
+		}
+	}
 
 	/* initialize all of the fd lists */
 	if (rte_memseg_list_walk(fd_list_create_walk, NULL))
-- 
2.17.1