Failed BUILD_BUG in mm/mremap.c

[nathanchance]

On next-20201208:

$ make -skj"$(nproc)" ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- LLVM=1 O=out distclean allmodconfig vmlinux
...
ld.lld: error: undefined symbol: __compiletime_assert_398
>>> referenced by mremap.c
>>>               mremap.o:(get_extent) in archive mm/built-in.a
...

get_extent:

/*
 * Returns an extent of the corresponding size for the pgt_entry specified if
 * valid. Else returns a smaller extent bounded by the end of the source and
 * destination pgt_entry.
 */
static unsigned long get_extent(enum pgt_entry entry, unsigned long old_addr,
			unsigned long old_end, unsigned long new_addr)
{
	unsigned long next, extent, mask, size;

	switch (entry) {
	case HPAGE_PMD:
	case NORMAL_PMD:
		mask = PMD_MASK;
		size = PMD_SIZE;
		break;
	case NORMAL_PUD:
		mask = PUD_MASK;
		size = PUD_SIZE;
		break;
	default:
		BUILD_BUG();
		break;
	}

	next = (old_addr + size) & mask;
	/* even if next overflowed, extent below will be ok */
	extent = (next > old_end) ? old_end - old_addr : next - old_addr;
	next = (new_addr + size) & mask;
	if (extent > next - new_addr)
		extent = next - new_addr;
	return extent;
}

Bisect points out this is probably UBSAN related:

$ git bisect log
# bad: [a9e26cb5f2615cd638f911ea96d4fff5b4d93690] Add linux-next specific files for 20201208
# good: [cd796ed3345030aa1bb332fe5c793b3dddaf56e7] Merge tag 'trace-v5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace
git bisect start 'a9e26cb5f2615cd638f911ea96d4fff5b4d93690' 'cd796ed3345030aa1bb332fe5c793b3dddaf56e7'
# good: [ad5711018462853bb0298d9c4caba35ee3e2a46e] Merge remote-tracking branch 'crypto/master'
git bisect good ad5711018462853bb0298d9c4caba35ee3e2a46e
# good: [6b415f03892954ad81976cd4025cf2afaa3d6ba7] Merge remote-tracking branch 'spi/for-next'
git bisect good 6b415f03892954ad81976cd4025cf2afaa3d6ba7
# good: [e4b95726aa03210eb2d74ba5fe1bd0d29f48c831] Merge remote-tracking branch 'staging/staging-next'
git bisect good e4b95726aa03210eb2d74ba5fe1bd0d29f48c831
# good: [0ba83260eb811bb3353d4437846e21a970e02a49] Merge remote-tracking branch 'userns/for-next'
git bisect good 0ba83260eb811bb3353d4437846e21a970e02a49
# good: [c0c70f8de1772385c98e4476e5c08bcfe5d4ea84] ath11k: make relay callbacks const
git bisect good c0c70f8de1772385c98e4476e5c08bcfe5d4ea84
# good: [bb87f5a0a0ef6f0c3553d3fec3249d3164dd533c] Merge remote-tracking branch 'auxdisplay/auxdisplay'
git bisect good bb87f5a0a0ef6f0c3553d3fec3249d3164dd533c
# bad: [1cfe18eb5ab53bf4086bd4afa7571a6eb412f615] epoll: replace gotos with a proper loop
git bisect bad 1cfe18eb5ab53bf4086bd4afa7571a6eb412f615
# good: [c26636d598da1890248ad157482e16c43b500ef9] scsi: block: fix for "scsi: block: Do not accept any requests while suspended"
git bisect good c26636d598da1890248ad157482e16c43b500ef9
# bad: [e0360dca9f06c66fe726724d571e883025329dc0] kmap: stupid hacks to make it compile
git bisect bad e0360dca9f06c66fe726724d571e883025329dc0
# bad: [f0350c9fad8a416a5797edb140e01bec21cc209e] ubsan: remove UBSAN_MISC in favor of individual options
git bisect bad f0350c9fad8a416a5797edb140e01bec21cc209e
# good: [995378db718490b00e56f0b749d3f1372c77c3c3] resource-fix-kernel-doc-markups-checkpatch-fixes
git bisect good 995378db718490b00e56f0b749d3f1372c77c3c3
# good: [2d60efe0a387aba744454ec2238b30285cf507b0] ubsan: disable object-size sanitizer under GCC
git bisect good 2d60efe0a387aba744454ec2238b30285cf507b0
# bad: [186c3e18dba3f035b73149f6356f8c1c439288f4] ubsan: enable for all*config builds
git bisect bad 186c3e18dba3f035b73149f6356f8c1c439288f4
# good: [a625b7b24fc318c6d731e941ce070a4d940ff143] ubsan: disable UBSAN_TRAP for all*config
git bisect good a625b7b24fc318c6d731e941ce070a4d940ff143
# first bad commit: [186c3e18dba3f035b73149f6356f8c1c439288f4] ubsan: enable for all*config builds

Sure enough, if I disable CONFIG_UBSAN_SANITIZE_ALL, the build passes just fine. I assume that this option prevents clang from eliminating the default statement? cc @kees

[nathanchance]

This is now in mainline.

[nathanchance]

Related to -fsanitize=unsigned-integer-overflow. cvise spits out:

enum { NORMAL_PMD } get_extent_old_addr;
get_extent_new_addr;
unsigned get_extent_next;
static get_extent(entry) {
  switch (entry) {
  case NORMAL_PMD:
    break;
  default:
    __compiletime_assert_1280();
  }
  get_extent_old_addr + --get_extent_next - get_extent_new_addr;
}
move_page_tables() { get_extent(); }

with this interestingness test.

[kees]

So this seems to be a UBSAN bug in Clang?

[kees]

I can't reproduce the failure.

builder@host:/srv/code$ cat mremap.i 
enum { NORMAL_PMD } get_extent_old_addr;
get_extent_new_addr;
unsigned get_extent_next;
static get_extent(entry) {
  switch (entry) {
  case NORMAL_PMD:
    break;
  default:
    __compiletime_assert_1280();
  }
  get_extent_old_addr + --get_extent_next - get_extent_new_addr;
}
move_page_tables() { get_extent(); }

builder@host:/srv/code$ clang-10 --version
Ubuntu clang version 10.0.1-6
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
builder@host:/srv/code$ rm mremap.o 
builder@host:/srv/code$ clang-10 -O2 -fsanitize=unsigned-integer-overflow -c -o mremap.o mremap.i 2>/dev/null
builder@host:/srv/code$ llvm-nm mremap.o 
                 U __ubsan_handle_add_overflow
                 U __ubsan_handle_sub_overflow
0000000000000004 C get_extent_new_addr
0000000000000004 C get_extent_next
0000000000000004 C get_extent_old_addr
0000000000000000 T move_page_tables

builder@host:/srv/code$ clang-12 --version
clang version 12.0.0 (https://github.com/llvm/llvm-project.git 0e94ba9d40d931fc1c647347302f39d24bd88d96)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /srv/llvm/x86/install/bin
builder@host:/srv/code$ rm mremap.o 
builder@host:/srv/code$ clang-12 -O2 -fsanitize=unsigned-integer-overflow -c -o mremap.o mremap.i 2>/dev/null
builder@host:/srv/code$ llvm-nm mremap.o 
                 U __ubsan_handle_add_overflow
                 U __ubsan_handle_sub_overflow
0000000000000004 B get_extent_new_addr
0000000000000008 B get_extent_next
0000000000000000 B get_extent_old_addr
0000000000000000 T move_page_tables

[nathanchance]

Huh odd... I will try to double check myself after work. Can you at least reproduce the failure with ARCH=arm64 allmodconfig vmlinux?

[kees]

Yeah, I can't even build x86. Same issue. I feel like I'm doing something wrong... :P

[nathanchance]

Hmmm cvise seems to have messed up... the final result fails the interestingness test :(

Oh well, used creduce to come up with:

enum { a } b;
c, d;
static e(g) {
  switch (g) {
  case a:
    break;
  default:
    __compiletime_assert_1280();
  }
  b + c - -b - d;
}
f() { e(); }

which reliably passes the interestingness test now.

[nickdesaulniers]

Causing issues for Android: https://android-review.googlesource.com/c/kernel/common/+/1534178

[nathanchance]

@arndb sent https://lore.kernel.org/lkml/20201230154104.522605-1-arnd@kernel.org/