On Tue, 2018-12-11 at 22:43 +0100, Arnd Bergmann wrote: > On Tue, Dec 11, 2018 at 6:18 PM Qian Cai <cai@xxxxxx> wrote: > > > > On Tue, 2018-12-11 at 13:42 +0100, Arnd Bergmann wrote: > > > On Fri, Dec 7, 2018 at 11:35 PM Qian Cai <cai@xxxxxx> wrote: > > > > > > > > If the kernel is configured with KASAN_EXTRA, the stack size is > > > > increasted significantly due to enable this option will set > > > > -fstack-reuse to "none" in GCC [1]. As the results, it could trigger > > > > stack overrun quite often with 32k stack size compiled using GCC 8. For > > > > example, this reproducer > > > > > > > > size > > > > 7536 shrink_inactive_list > > > > 7440 shrink_page_list > > > > 6560 fscache_stats_show > > > > 3920 jbd2_journal_commit_transaction > > > > 3216 try_to_unmap_one > > > > 3072 migrate_page_move_mapping > > > > 3584 migrate_misplaced_transhuge_page > > > > 3920 ip_vs_lblcr_schedule > > > > 4304 lpfc_nvme_info_show > > > > 3888 lpfc_debugfs_nvmestat_data.constprop > > > > > > > > There are other 49 functions are over 2k in size while compiling kernel > > > > with "-Wframe-larger-than=" on this machine. Hence, it is too much work > > > > to change Makefiles for each object to compile without > > > > -fsanitize-address-use-after-scope individually. > > > > > > > > [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81715#c23 > > > > > > Could you clarify: are the numbers you see with or without the bugfix > > > from that bugzilla? > > > > > > > The numbers were from GCC8 which does NOT contain this patch [1]. > > > > GCC9 is awesome which reduced the numbers in half even for KASAN_EXTRA. Only > > thing is that GCC9 has not been officially released yet, so it is a bit > > inconvenient for users need to compile the compiler by themselves first. > > > > I am fine either way to drop this patch or keep it until GCC9 is GA. > > > > [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81715#c35 > > Maybe we can make the constant depend on the compiler version? I am not too keen to do the version-check considering some LTS versions could just back-port those patches and the render the version-check incorrectly. > It may also be possible to reduce the KASAN_THREAD_SHIFT > constant for the normal case with gcc-9 and go back to the > default frame size then.
