https://bugs.kde.org/show_bug.cgi?id=526974
--- Comment #1 from Martin Cermak <[email protected]> --- Created attachment 197076 --> https://bugs.kde.org/attachment.cgi?id=197076&action=edit proposed patch against LTP The mremap03 test underwent a rewrite comparing ltp-full-20260529 and ltp-full-20260930. The LTP commit is: -------------------------------------------------8<--------------------------------------------------------------- commit 88cfde5fbc14b23f673b81855127448a519679c0 Author: Andrea Cervesato <[email protected]> Date: Tue Jul 21 10:37:50 2026 +0200 mremap03: Convert to new API Convert the mremap03 test case from the legacy LTP API to the new tst_test API. Rewrite the test to use TST_EXP_FAIL_PTR_VOID() to verify that mremap() fails with MAP_FAILED and EFAULT when growing a mapping whose old_address refers to an unmapped region obtained via tst_get_bad_addr(). -------------------------------------------------8<--------------------------------------------------------------- >From the functional persp., the difference is in it's setup() function. The original version prepares the bad_addr like this: bad_addr = tst_get_bad_addr(cleanup); The rewritten version prepares the bad addr like this: bad_addr = tst_get_bad_addr(NULL); SAFE_MUNMAP(bad_addr, page_size); The different arg (cleanup vs. NULL) isn't important in case mmap() inside tst_get_bad_addr() passes. That said, the original version generates a mapped address which is inaccessible (PROT_NONE). The rewritten version generates the same kind of bad address, but unmaps it right away. Passing the unmapped bad_addr to to mremap() makes valgrind unhappy. Such bad_addr seems to be too bad :) Commenting out the SAFE_MUNMAP() in the rewritten version of the test make valgrind happy again. It doesn't break the testcase either. That said, I suspect there is a bug in the LTP test and the attached patch fixes it. -- You are receiving this mail because: You are watching all bug changes.
