#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>
#include <rte_eal_memconfig.h>
#include <rte_eal.h>
+#include <rte_errno.h>
#include <rte_memory.h>
#include <rte_spinlock.h>
#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
+
+/*
+ * 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.
+ */
+static 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
*/
static int fallocate_supported = -1; /* unknown */
-/* for single-file segments, we need some kind of mechanism to keep track of
+/*
+ * we have two modes - single file segments, and file-per-page mode.
+ *
+ * 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.
+ * bunch of lockfiles for each page. so, we will use 'fds' array to keep track
+ * of per-page lockfiles. we will store the actual segment list fd in the
+ * 'memseg_list_fd' field.
+ *
+ * for file-per-page mode, each page will have its own fd, so 'memseg_list_fd'
+ * will be invalid (set to -1), and we'll use 'fds' to keep track of page fd's.
*
* 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.
+ * segments.
*/
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];
+} fd_list[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];
}
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);
return st.st_size;
}
-/* 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 inline uint32_t
+bsf64(uint64_t v)
{
- struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
- int i, ret = 0;
-
- for (i = 0; i < RTE_MAX_MEMSEG_LISTS; i++) {
- struct rte_memseg_list *msl = &mcfg->memsegs[i];
+ return (uint32_t)__builtin_ctzll(v);
+}
- if (msl->base_va == NULL)
- continue;
+static inline uint32_t
+log2_u64(uint64_t v)
+{
+ if (v == 0)
+ return 0;
+ v = rte_align64pow2(v);
+ return bsf64(v);
+}
- ret = func(msl, arg);
- if (ret < 0)
- return -1;
- if (ret > 0)
- return 1;
- }
- return 0;
+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 */
char path[PATH_MAX] = {0};
int fd;
- if (list_idx < 0 || list_idx >= (int)RTE_DIM(lock_fds))
+ if (list_idx < 0 || list_idx >= (int)RTE_DIM(fd_list))
return -1;
- if (seg_idx < 0 || seg_idx >= lock_fds[list_idx].len)
+ if (seg_idx < 0 || seg_idx >= fd_list[list_idx].len)
return -1;
- fd = lock_fds[list_idx].fds[seg_idx];
+ fd = fd_list[list_idx].fds[seg_idx];
/* does this lock already exist? */
if (fd >= 0)
return fd;
return -1;
}
/* store it for future reference */
- lock_fds[list_idx].fds[seg_idx] = fd;
- lock_fds[list_idx].count++;
+ fd_list[list_idx].fds[seg_idx] = fd;
+ fd_list[list_idx].count++;
return fd;
}
{
int fd, ret;
- if (list_idx < 0 || list_idx >= (int)RTE_DIM(lock_fds))
+ if (list_idx < 0 || list_idx >= (int)RTE_DIM(fd_list))
return -1;
- if (seg_idx < 0 || seg_idx >= lock_fds[list_idx].len)
+ if (seg_idx < 0 || seg_idx >= fd_list[list_idx].len)
return -1;
- fd = lock_fds[list_idx].fds[seg_idx];
+ fd = fd_list[list_idx].fds[seg_idx];
/* upgrade lock to exclusive to see if we can remove the lockfile */
ret = lock(fd, LOCK_EX);
* and remove it from list anyway.
*/
close(fd);
- lock_fds[list_idx].fds[seg_idx] = -1;
- lock_fds[list_idx].count--;
+ fd_list[list_idx].fds[seg_idx] = -1;
+ fd_list[list_idx].count--;
if (ret < 0)
return -1;
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);
- fd = lock_fds[list_idx].memseg_list_fd;
+ fd = fd_list[list_idx].memseg_list_fd;
if (fd < 0) {
fd = open(path, O_CREAT | O_RDWR, 0600);
close(fd);
return -1;
}
- lock_fds[list_idx].memseg_list_fd = fd;
+ fd_list[list_idx].memseg_list_fd = fd;
}
} else {
/* 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);
+
+ fd = fd_list[list_idx].fds[seg_idx];
+
if (fd < 0) {
- RTE_LOG(DEBUG, EAL, "%s(): open failed: %s\n", __func__,
- 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;
+ fd = open(path, O_CREAT | O_RDWR, 0600);
+ if (fd < 0) {
+ RTE_LOG(DEBUG, EAL, "%s(): open failed: %s\n",
+ __func__, 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;
+ }
+ fd_list[list_idx].fds[seg_idx] = fd;
}
}
return fd;
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
* 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) {
+ if (ret < 1 && fd_list[list_idx].count == 0) {
close(fd);
- lock_fds[list_idx].memseg_list_fd = -1;
+ fd_list[list_idx].memseg_list_fd = -1;
}
if (ret < 0) {
* more segments active in this segment list,
* and remove the file if there aren't.
*/
- if (lock_fds[list_idx].count == 0) {
+ if (fd_list[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;
+ fd_list[list_idx].memseg_list_fd = -1;
}
}
}
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;
+ int flags;
+ void *new_addr;
- /* 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");
+ alloc_sz = hi->hugepage_sz;
+
+ /* 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;
}
- alloc_sz = hi->hugepage_sz;
- 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 {
+ /* 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;
+ mmap_flags = flags;
+
+ /* 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;
- 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;
+ }
+
+ 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 &&
+ !internal_config.in_memory) {
+ if (unlink(path)) {
+ RTE_LOG(DEBUG, EAL, "%s(): unlink() failed: %s\n",
+ __func__, strerror(errno));
+ goto resized;
+ }
+ }
}
+ 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.
+ * 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);
+ 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));
- 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__);
goto resized;
}
- rte_iova_t iova = rte_mem_virt2iova(addr);
- if (iova == RTE_BAD_PHYS_ADDR) {
- RTE_LOG(DEBUG, EAL, "%s(): can't get IOVA addr\n",
- __func__);
- goto mapped;
- }
-
-#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES
- move_pages(getpid(), 1, &addr, NULL, &cur_socket_id, 0);
-
- if (cur_socket_id != socket_id) {
- RTE_LOG(DEBUG, EAL,
- "%s(): allocation happened on wrong socket (wanted %d, got %d)\n",
- __func__, socket_id, cur_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
(unsigned int)(alloc_sz >> 20));
goto mapped;
}
- /* for non-single file segments, we can close fd here */
- if (!internal_config.single_file_segments)
- close(fd);
/* 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
*/
*(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__);
+ goto mapped;
+ }
+
+#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES
+ move_pages(getpid(), 1, &addr, NULL, &cur_socket_id, 0);
+
+ if (cur_socket_id != socket_id) {
+ RTE_LOG(DEBUG, EAL,
+ "%s(): allocation happened on wrong socket (wanted %d, got %d)\n",
+ __func__, socket_id, cur_socket_id);
+ goto mapped;
+ }
+#endif
+
ms->addr = addr;
ms->hugepage_sz = alloc_sz;
ms->len = alloc_sz;
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:
+ /* some codepaths will return negative fd, so exit early */
+ if (fd < 0)
+ return -1;
+
if (internal_config.single_file_segments) {
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 (lock(fd, LOCK_EX) == 1)
+ if (internal_config.hugepage_unlink == 0 &&
+ internal_config.in_memory == 0 &&
+ lock(fd, LOCK_EX) == 1)
unlink(path);
close(fd);
+ fd_list[list_idx].fds[seg_idx] = -1;
}
return -1;
}
{
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);
- /* if we are not in single file segments mode, we're going to unmap the
- * segment and thus drop the lock on original fd, so take out another
- * shared lock before we do that.
- */
- fd = get_seg_fd(path, sizeof(path), hi, list_idx, seg_idx);
- if (fd < 0)
- return -1;
-
if (mmap(ms->addr, ms->len, PROT_READ,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0) ==
MAP_FAILED) {
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.in_memory && internal_config.hugepage_unlink)
+ exit_early = true;
+
+ if (exit_early) {
+ 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, path, list_idx, seg_idx, map_offset,
/* 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);
+ fd_list[list_idx].fds[seg_idx] = -1;
}
memset(ms, 0, sizeof(*ms));
- return ret;
+ return ret < 0 ? -1 : 0;
}
struct alloc_walk_param {
struct alloc_walk_param *wa = arg;
struct rte_memseg_list *cur_msl;
size_t page_sz;
- int cur_idx, start_idx, j, dir_fd;
+ int cur_idx, start_idx, j, dir_fd = -1;
unsigned int msl_idx, need, i;
if (msl->page_sz != wa->page_sz)
* 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.
*/
- 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;
+ 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++) {
&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);
- close(dir_fd);
+ if (dir_fd >= 0)
+ close(dir_fd);
return -1;
}
if (wa->ms)
wa->segs_allocated = i;
if (i > 0)
cur_msl->version++;
- close(dir_fd);
+ if (dir_fd >= 0)
+ close(dir_fd);
return 1;
}
struct rte_memseg_list *found_msl;
struct free_walk_param *wa = arg;
uintptr_t start_addr, end_addr;
- int msl_idx, seg_idx, ret, dir_fd;
+ 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;
+ end_addr = start_addr + msl->len;
if ((uintptr_t)wa->ms->addr < start_addr ||
(uintptr_t)wa->ms->addr >= end_addr)
* 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.
*/
- 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;
+ 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++;
ret = free_seg(wa->ms, wa->hi, msl_idx, seg_idx);
- close(dir_fd);
+ if (dir_fd >= 0)
+ close(dir_fd);
if (ret < 0)
return -1;
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__);
#ifdef RTE_EAL_NUMA_AWARE_HUGEPAGES
if (have_numa)
- resotre_numa(&oldpolicy, oldmask);
+ restore_numa(&oldpolicy, oldmask);
#endif
return ret;
}
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)
unsigned int i;
int msl_idx;
+ if (msl->external)
+ return 0;
+
msl_idx = msl - mcfg->memsegs;
primary_msl = &mcfg->memsegs[msl_idx];
local_msl = &local_memsegs[msl_idx];
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;
}
char name[PATH_MAX];
int msl_idx, ret;
+ if (msl->external)
+ return 0;
+
msl_idx = msl - mcfg->memsegs;
primary_msl = &mcfg->memsegs[msl_idx];
local_msl = &local_memsegs[msl_idx];
return -1;
}
local_msl->base_va = primary_msl->base_va;
+ local_msl->len = primary_msl->len;
return 0;
}
static int
-secondary_lock_list_create_walk(const struct rte_memseg_list *msl,
- void *arg __rte_unused)
+alloc_list(int list_idx, int len)
{
- struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
- unsigned int i, len;
- int msl_idx;
int *data;
+ int i;
- msl_idx = msl - mcfg->memsegs;
- len = msl->memseg_arr.len;
-
- /* ensure we have space to store lock fd per each possible segment */
+ /* ensure we have space to store 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");
+ RTE_LOG(ERR, EAL, "Unable to allocate space for file 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;
+ fd_list[list_idx].fds = data;
+ fd_list[list_idx].len = len;
+ fd_list[list_idx].count = 0;
+ fd_list[list_idx].memseg_list_fd = -1;
+
+ return 0;
+}
+
+static int
+fd_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 len;
+ int msl_idx;
+
+ if (msl->external)
+ return 0;
+
+ msl_idx = msl - mcfg->memsegs;
+ len = msl->memseg_arr.len;
+
+ return alloc_list(msl_idx, len);
+}
+
+int
+eal_memalloc_set_seg_fd(int list_idx, int seg_idx, int fd)
+{
+ struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+
+ /* if list is not allocated, allocate it */
+ if (fd_list[list_idx].len == 0) {
+ int len = mcfg->memsegs[list_idx].memseg_arr.len;
+
+ if (alloc_list(list_idx, len) < 0)
+ return -ENOMEM;
+ }
+ fd_list[list_idx].fds[seg_idx] = fd;
+
+ return 0;
+}
+
+int
+eal_memalloc_get_seg_fd(int list_idx, int seg_idx)
+{
+ int fd;
+ if (internal_config.single_file_segments) {
+ fd = fd_list[list_idx].memseg_list_fd;
+ } else if (fd_list[list_idx].len == 0) {
+ /* list not initialized */
+ fd = -1;
+ } else {
+ fd = fd_list[list_idx].fds[seg_idx];
+ }
+ if (fd < 0)
+ return -ENODEV;
+ return fd;
+}
+
+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_get_seg_fd_offset(int list_idx, int seg_idx, size_t *offset)
+{
+ struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
+ /* fd_list not initialized? */
+ if (fd_list[list_idx].len == 0)
+ return -ENODEV;
+ if (internal_config.single_file_segments) {
+ size_t pgsz = mcfg->memsegs[list_idx].page_sz;
+
+ /* segment not active? */
+ if (fd_list[list_idx].memseg_list_fd < 0)
+ return -ENOENT;
+ *offset = pgsz * seg_idx;
+ } else {
+ /* segment not active? */
+ if (fd_list[list_idx].fds[seg_idx] < 0)
+ return -ENOENT;
+ *offset = 0;
+ }
return 0;
}
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();
- /* initialize all of the lock fd lists */
- if (internal_config.single_file_segments)
- if (rte_memseg_list_walk(secondary_lock_list_create_walk, NULL))
+ 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))
+ return -1;
return 0;
}