This is an automated email from the ASF dual-hosted git repository.

xiaoxiang781216 pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/nuttx.git

commit 9e0390259ad7771c0cb54d1d06c952cd3faff78d
Author: dengwenqi <[email protected]>
AuthorDate: Sat Oct 3 14:30:44 2026 +0800

    backtrace: skip KASAN instrumentation in backtrace_remove to fix busyloop
    
    backtrace_remove() runs whenever a backtrace record is released.  On
    KASAN-enabled debug builds its loads of entry->ref, the bucket slot and
    the collision chain each trip a __asan_loadN_noabort shadow check,
    which adds up when records are released at a high rate, e.g. for every
    free during a recursive directory scan.  backtrace_hash(),
    backtrace_find() and backtrace_alloc() on the same path already skip
    the instrumentation.
    
    Add nosanitize_address to backtrace_remove().  All pointers it touches
    are pool-internal (entry, bucket, freelist nodes), so skipping the
    shadow check is safe.
    
    Signed-off-by: dengwenqi <[email protected]>
---
 libs/libc/misc/lib_backtrace.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/libs/libc/misc/lib_backtrace.c b/libs/libc/misc/lib_backtrace.c
index fd53fbdad93..e4f697acaf1 100644
--- a/libs/libc/misc/lib_backtrace.c
+++ b/libs/libc/misc/lib_backtrace.c
@@ -493,6 +493,18 @@ FAR void *backtrace_record(int skip)
  *   Return 0 if success, otherwise return a negative value.
  ****************************************************************************/
 
+/* nosanitize_address: backtrace_remove is on the hot path of every
+ * allocation/free (mempool_record calls it for each when
+ * CONFIG_MM_RECORD_STACK is on).  On KASAN-enabled debug builds the loads
+ * of entry->ref, the bucket slot and the collision chain each trip a
+ * per-access shadow check; across a heavy allocation phase (e.g. recursive
+ * directory scan at boot) this pushes wall time past the test framework's
+ * busy-loop threshold.  All pointers touched are pool-internal, so skipping
+ * the check is safe -- same reasoning as backtrace_hash()/backtrace_find();
+ * this is the remaining uninstrumented accessor on the same path.
+ */
+
+nosanitize_address
 void backtrace_remove(FAR void *index)
 {
   FAR struct backtrace_pool_s *bp = &g_backtrace_pool;

Reply via email to