When running with --in-memory, EAL uses anonymous hugepages via
memfd_create() and never needs hugetlbfs files.

However, hugepage_info_init() only skipped the mountpoint and
directory locking when no hugetlbfs mount was found.
When a mount existed, the code fell through to open and flock()
the hugepage directory regardless of --in-memory.

This causes initialization failures when EAL encounters a hugetlbfs
mount it cannot access (e.g., mounts owned by another component on
the system), even though in-memory mode does not need the mount at all.

Skip the mountpoint lookup and directory locking entirely when
--in-memory is set.
The hugepage sizes and counts are read from sysfs, which is all
that in-memory mode needs.

Fixes: 72b49ff623c4 ("mem: support --in-memory mode")
Cc: [email protected]

Signed-off-by: David Marchand <[email protected]>
---
 lib/eal/linux/eal_hugepage_info.c | 30 ++++++++++++++----------------
 1 file changed, 14 insertions(+), 16 deletions(-)

diff --git a/lib/eal/linux/eal_hugepage_info.c 
b/lib/eal/linux/eal_hugepage_info.c
index 05c5b3f613..781eb3787b 100644
--- a/lib/eal/linux/eal_hugepage_info.c
+++ b/lib/eal/linux/eal_hugepage_info.c
@@ -480,6 +480,20 @@ hugepage_info_init(void)
                hpi->hugepage_sz =
                        rte_str_to_size(&dirent->d_name[dirent_start_len]);
 
+#ifdef MAP_HUGE_SHIFT
+               /* if we have kernel support for reserving hugepages
+                * through mmap, and we're in in-memory mode, treat this
+                * page size as valid. we cannot be in legacy mode at
+                * this point because we've checked this earlier in the
+                * init process.
+                */
+               if (internal_conf->in_memory) {
+                       calc_num_pages(hpi, dirent, 0);
+                       num_sizes++;
+                       continue;
+               }
+#endif
+
                /* first, check if we have a mountpoint */
                if (get_hugepage_dir(hpi->hugepage_sz,
                        hpi->hugedir, sizeof(hpi->hugedir)) < 0) {
@@ -493,22 +507,6 @@ hugepage_info_init(void)
                                        "%" PRIu64 " reserved, but no mounted "
                                        "hugetlbfs found for that size",
                                        num_pages, hpi->hugepage_sz);
-                       /* if we have kernel support for reserving hugepages
-                        * through mmap, and we're in in-memory mode, treat this
-                        * page size as valid. we cannot be in legacy mode at
-                        * this point because we've checked this earlier in the
-                        * init process.
-                        */
-#ifdef MAP_HUGE_SHIFT
-                       if (internal_conf->in_memory) {
-                               EAL_LOG(DEBUG, "In-memory mode enabled, "
-                                       "hugepages of size %" PRIu64 " bytes "
-                                       "will be allocated anonymously",
-                                       hpi->hugepage_sz);
-                               calc_num_pages(hpi, dirent, 0);
-                               num_sizes++;
-                       }
-#endif
                        continue;
                }
 
-- 
2.56.1

Reply via email to