Re: [lkp] [Btrfs] e5a7b11a5c: shift exponent 32 is too large for 32-bit type 'int'

[Omar Sandoval]

On Fri, Jul 15, 2016 at 09:51:20AM +0800, kernel test robot wrote:
> 
> FYI, we noticed the following commit:
> 
> https://github.com/0day-ci/linux Omar-Sandoval/Btrfs-fix-extent-buffer-bitmap-tests-on-big-endian-systems/20160713-074511
> commit e5a7b11a5caa8d42fe3fb4ff2d75b6811aeb8eb7 ("Btrfs: fix extent buffer bitmap tests on big-endian systems")
> 
> in testcase: boot
> 
> on test machine: 2 threads qemu-system-x86_64 -enable-kvm -cpu Haswell,+smep,+smap with 1G memory
> 
> caused below changes:
> 
> 
> +------------------------------------------------------------------+----------+------------+
> |                                                                  | v4.7-rc7 | e5a7b11a5c |
> +------------------------------------------------------------------+----------+------------+
> | boot_successes                                                   | 0        | 0          |
> | boot_failures                                                    | 80       | 8          |
> | invoked_oom-killer:gfp_mask=0x                                   | 59       | 2          |
> | Mem-Info                                                         | 59       | 2          |
> | Kernel_panic-not_syncing:Out_of_memory_and_no_killable_processes | 59       | 2          |
> | backtrace:btrfs_test_extent_io                                   | 35       | 6          |
> | backtrace:init_btrfs_fs                                          | 35       | 6          |
> | backtrace:kernel_init_freeable                                   | 59       | 8          |
> | BUG:kernel_test_oversize                                         | 11       | 4          |
> | backtrace:vfs_write                                              | 24       | 2          |
> | backtrace:SyS_write                                              | 24       | 2          |
> | backtrace:populate_rootfs                                        | 24       | 2          |
> | kernel_BUG_at_mm/filemap.c                                       | 1        |            |
> | invalid_opcode:#[##]PREEMPT_SMP                                  | 1        |            |
> | RIP:unlock_page                                                  | 1        |            |
> | Kernel_panic-not_syncing:Fatal_exception                         | 1        |            |
> | backtrace:kswapd                                                 | 1        |            |
> | BUG:kernel_boot_hang                                             | 9        |            |
> | BUG:kernel_torture_test_oversize                                 | 0        | 2          |
> +------------------------------------------------------------------+----------+------------+
> 
> 
> 
> [   74.703186] UBSAN: Undefined behaviour in fs/btrfs/tests/extent-io-tests.c:367:4
> [   74.703359] shift exponent 32 is too large for 32-bit type 'int'
> [   74.703359] CPU: 1 PID: 1 Comm: swapper/0 Not tainted 4.7.0-rc7-00001-ge5a7b11 #1
> [   74.703359] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
> [   74.703359]  00000000ffffffff 0000000000000097 0000000000000001 0000000000000001
> [   74.703359]  0000000000000000 0000000000000001 ffff880035143c48 ffffffff82046359
> [   74.703359]  ffffffff813e786b 000000007a4be528 000000007a4be500 0000000000000020
> [   74.703359] Call Trace:
> [   74.703359]  [<ffffffff82046359>] dump_stack+0x290/0x3f0
> [   74.703359]  [<ffffffff813e786b>] ? printk+0x5d/0x65
> [   74.703359]  [<ffffffff8212cbd9>] ubsan_epilogue+0x12/0x3f
> [   74.703359]  [<ffffffff8212d61b>] __ubsan_handle_shift_out_of_bounds+0x19c/0x1d1
> [   74.703359]  [<ffffffff81ca5a42>] ? extent_buffer_bitmap_set+0x508/0x517
> [   74.703359]  [<ffffffff81e142e6>] __test_eb_bitmaps+0x5f1/0x7a7
> [   74.703359]  [<ffffffff81e142e6>] ? __test_eb_bitmaps+0x5f1/0x7a7
> [   74.703359]  [<ffffffff81e16479>] btrfs_test_extent_io+0x281/0x427
> [   74.703359]  [<ffffffff81d2b075>] ? cleanup_write_cache_enospc+0x88/0x88
> [   74.703359]  [<ffffffff8606d299>] init_btrfs_fs+0x926/0xd88
> [   74.703359]  [<ffffffff8606c973>] ? trace_event_define_fields_btrfs_space_reservation+0x32f/0x32f
> [   74.703359]  [<ffffffff81000884>] do_one_initcall+0x240/0x4ed
> [   74.703359]  [<ffffffff85fd107c>] kernel_init_freeable+0x3b7/0x628
> [   74.703359]  [<ffffffff83033b7f>] kernel_init+0x1c/0x49c
> [   74.703359]  [<ffffffff830600cf>] ret_from_fork+0x1f/0x40
> [   74.703359]  [<ffffffff83033b63>] ? rest_init+0x2aa/0x2aa
> [   74.703359] ================================================================================
> [   75.122581] BTRFS: selftest: Extent I/O tests finished
> [   75.123550] BTRFS: selftest: Running btrfs_get_extent tests

Hm, that doesn't make much sense:

365         for (i = 0; i < len * BITS_PER_BYTE / 32; i++) {
366                 x = (0x19660dULL * (u64)x + 0x3c6ef35fULL) & 0xffffffffU;
367                 for (j = 0; j < 32; j++) {
368                         if (x & (1U << j)) {
369                                 bitmap_set(bitmap, i * 32 + j, 1);
370                                 extent_buffer_bitmap_set(eb, 0, i * 32 + j, 1);
371                         }
372                 }
373         }

How are we shifting by 32? I'll try to reproduce this tomorrow, but I
wonder if it's a compiler/UBSAN bug.

-- 
Omar