Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Stack's lower bits are randomized despite ADDR_NO_RANDOMIZE #52

Closed
XeR opened this issue Sep 22, 2020 · 1 comment
Closed

Stack's lower bits are randomized despite ADDR_NO_RANDOMIZE #52

XeR opened this issue Sep 22, 2020 · 1 comment
Labels
bug Something isn't working good first issue Good for newcomers

Comments

@XeR
Copy link

XeR commented Sep 22, 2020

Issue

The stack's lower bits are randomized when randomize_va_space is not 0
regardless of the ADDR_NO_RANDOMIZE personality.

This is inconvenient when debugging software. It requires to turn off
ASLR on the whole system to have deterministic stack addresses.

Test

This can be tested with the following code :

void main()
{
	int a;
	printf("%p\n", &a);
}

On linux-hardened

$ for i in {1..5}; do ./a.out; done
0x73a0feb8e134
0x7d5c898ca584
0x7f2a18026ae4
0x7ff12a7c2e14
0x791eeca04c04
$ for i in {1..5}; do setarch $(uname -m) -R ./a.out; done
0x7fffffffd204
0x7fffffffd304
0x7fffffffde24
0x7fffffffd4c4
0x7fffffffd564
# sysctl -w kernel.randomize_va_space=0
$ for i in {1..5}; do ./a.out; done
0x7fffffffe084
0x7fffffffe084
0x7fffffffe084
0x7fffffffe084
0x7fffffffe084

On a vanilla kernel

$ for i in {1..5}; do ./a.out; done
0x7ffe5ed5cc3c
0x7ffefeaca9bc
0x7ffe8e5b28bc
0x7ffd8144582c
0x7ffd4a0247ec
$ for i in {1..5}; do setarch $(uname -m) -R ./a.out; done`
0x7fffffffe42c
0x7fffffffe42c
0x7fffffffe42c
0x7fffffffe42c
0x7fffffffe42c

Root cause

setup_arg_pages in fs/exec.c calls arch_align_stack (unless
CONFIG_STACK_GROWS_UP is set).

arch_align_stack checks both randomize_va_space and ADDR_NO_RANDOMIZE :

if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
	sp -= get_random_int() % 8192;

This results in the lower bits of the stack being randomized in both linux and
linux-hardened.

Commit 533d2e0 randomizes the stack a second
time. This code, only checks for randomize_va_space:

if (randomize_va_space)
	bprm->p ^= get_random_int() & ~PAGE_MASK;

I assume this commit exists since arch_align_stack is not defined for every
architecture. Only x86, aarch64, ppc, s390, mips and um. (most notably: arm
does not)

cf. arch/um/kernel/process.c:

/*
 * Only x86 and x86_64 have an arch_align_stack().
 * All other arches have "#define arch_align_stack(x) (x)"
 * in their asm/exec.h
 */

Can we check the ADDR_NO_RANDOMIZE personality before xoring the stack pointer
with a random value?

@anthraxx anthraxx added bug Something isn't working good first issue Good for newcomers labels Dec 22, 2020
@anthraxx
Copy link
Owner

fixed in 6b2d22e

anthraxx pushed a commit that referenced this issue Mar 13, 2021
commit 64b9cea upstream.

Syzkaller was able to trigger the following splat again:

WARNING: CPU: 1 PID: 12512 at net/mptcp/protocol.c:761 mptcp_reset_timer+0x12a/0x160 net/mptcp/protocol.c:761
Modules linked in:
CPU: 1 PID: 12512 Comm: kworker/1:6 Not tainted 5.10.0-rc6 #52
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
Workqueue: events mptcp_worker
RIP: 0010:mptcp_reset_timer+0x12a/0x160 net/mptcp/protocol.c:761
Code: e8 4b 0c ad ff e8 56 21 88 fe 48 b8 00 00 00 00 00 fc ff df 48 c7 04 03 00 00 00 00 48 83 c4 40 5b 5d 41 5c c3 e8 36 21 88 fe <0f> 0b 41 bc c8 00 00 00 eb 98 e8 e7 b1 af fe e9 30 ff ff ff 48 c7
RSP: 0018:ffffc900018c7c68 EFLAGS: 00010293
RAX: ffff888108cb1c80 RBX: 1ffff92000318f8d RCX: ffffffff82ad0307
RDX: 0000000000000000 RSI: ffffffff82ad036a RDI: 0000000000000007
RBP: ffff888113e2d000 R08: ffff888108cb1c80 R09: ffffed10227c5ab7
R10: ffff888113e2d5b7 R11: ffffed10227c5ab6 R12: 0000000000000000
R13: ffff88801f100000 R14: ffff888113e2d5b0 R15: 0000000000000001
FS:  0000000000000000(0000) GS:ffff88811b500000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fd76a874ef8 CR3: 000000001689c005 CR4: 0000000000170ee0
Call Trace:
 mptcp_worker+0xaa4/0x1560 net/mptcp/protocol.c:2334
 process_one_work+0x8d3/0x1200 kernel/workqueue.c:2272
 worker_thread+0x9c/0x1090 kernel/workqueue.c:2418
 kthread+0x303/0x410 kernel/kthread.c:292
 ret_from_fork+0x22/0x30 arch/x86/entry/entry_64.S:296

The mptcp_worker tries to update the MPTCP retransmission timer
even if such timer is not currently scheduled.

The mptcp_rtx_head() return value is bogus: we can have enqueued
data not yet transmitted. The above may additionally cause spurious,
unneeded MPTCP-level retransmissions.

Fix the issue adding an explicit clearing of the rtx queue before
trying to retransmit and checking for unacked data.
Additionally drop an unneeded timer stop call and the unused
mptcp_rtx_tail() helper.

Reported-by: Christoph Paasch <cpaasch@apple.com>
Fixes: 6e628cd ("mptcp: use mptcp release_cb for delayed tasks")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Mat Martineau <mathew.j.martineau@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
anthraxx pushed a commit that referenced this issue May 23, 2021
[ Upstream commit d5027ca ]

Ritesh reported a bug [1] against UML, noting that it crashed on
startup. The backtrace shows the following (heavily redacted):

(gdb) bt
...
 #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268
 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2
 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72
...
 #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359
...
 #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486
 #45 0x00007f8990968b85 in __getgrnam_r [...]
 #46 0x00007f89909d6b77 in grantpt [...]
 #47 0x00007f8990a9394e in __GI_openpty [...]
 #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407
 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598
 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45
 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334
 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144

indicating that the UML function openpty_cb() calls openpty(),
which internally calls __getgrnam_r(), which causes the nsswitch
machinery to get started.

This loads, through lots of indirection that I snipped, the
libcom_err.so.2 library, which (in an unknown function, "??")
calls sem_init().

Now, of course it wants to get libpthread's sem_init(), since
it's linked against libpthread. However, the dynamic linker
looks up that symbol against the binary first, and gets the
kernel's sem_init().

Hajime Tazaki noted that "objcopy -L" can localize a symbol,
so the dynamic linker wouldn't do the lookup this way. I tried,
but for some reason that didn't seem to work.

Doing the same thing in the linker script instead does seem to
work, though I cannot entirely explain - it *also* works if I
just add "VERSION { { global: *; }; }" instead, indicating that
something else is happening that I don't really understand. It
may be that explicitly doing that marks them with some kind of
empty version, and that's different from the default.

Explicitly marking them with a version breaks kallsyms, so that
doesn't seem to be possible.

Marking all the symbols as local seems correct, and does seem
to address the issue, so do that. Also do it for static link,
nsswitch libraries could still be loaded there.

[1] https://bugs.debian.org/983379

Reported-by: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Tested-By: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue May 23, 2021
[ Upstream commit d5027ca ]

Ritesh reported a bug [1] against UML, noting that it crashed on
startup. The backtrace shows the following (heavily redacted):

(gdb) bt
...
 #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268
 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2
 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72
...
 #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359
...
 #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486
 #45 0x00007f8990968b85 in __getgrnam_r [...]
 #46 0x00007f89909d6b77 in grantpt [...]
 #47 0x00007f8990a9394e in __GI_openpty [...]
 #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407
 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598
 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45
 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334
 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144

indicating that the UML function openpty_cb() calls openpty(),
which internally calls __getgrnam_r(), which causes the nsswitch
machinery to get started.

This loads, through lots of indirection that I snipped, the
libcom_err.so.2 library, which (in an unknown function, "??")
calls sem_init().

Now, of course it wants to get libpthread's sem_init(), since
it's linked against libpthread. However, the dynamic linker
looks up that symbol against the binary first, and gets the
kernel's sem_init().

Hajime Tazaki noted that "objcopy -L" can localize a symbol,
so the dynamic linker wouldn't do the lookup this way. I tried,
but for some reason that didn't seem to work.

Doing the same thing in the linker script instead does seem to
work, though I cannot entirely explain - it *also* works if I
just add "VERSION { { global: *; }; }" instead, indicating that
something else is happening that I don't really understand. It
may be that explicitly doing that marks them with some kind of
empty version, and that's different from the default.

Explicitly marking them with a version breaks kallsyms, so that
doesn't seem to be possible.

Marking all the symbols as local seems correct, and does seem
to address the issue, so do that. Also do it for static link,
nsswitch libraries could still be loaded there.

[1] https://bugs.debian.org/983379

Reported-by: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Tested-By: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue May 23, 2021
[ Upstream commit d5027ca ]

Ritesh reported a bug [1] against UML, noting that it crashed on
startup. The backtrace shows the following (heavily redacted):

(gdb) bt
...
 #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268
 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2
 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72
...
 #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359
...
 #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486
 #45 0x00007f8990968b85 in __getgrnam_r [...]
 #46 0x00007f89909d6b77 in grantpt [...]
 #47 0x00007f8990a9394e in __GI_openpty [...]
 #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407
 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598
 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45
 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334
 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144

indicating that the UML function openpty_cb() calls openpty(),
which internally calls __getgrnam_r(), which causes the nsswitch
machinery to get started.

This loads, through lots of indirection that I snipped, the
libcom_err.so.2 library, which (in an unknown function, "??")
calls sem_init().

Now, of course it wants to get libpthread's sem_init(), since
it's linked against libpthread. However, the dynamic linker
looks up that symbol against the binary first, and gets the
kernel's sem_init().

Hajime Tazaki noted that "objcopy -L" can localize a symbol,
so the dynamic linker wouldn't do the lookup this way. I tried,
but for some reason that didn't seem to work.

Doing the same thing in the linker script instead does seem to
work, though I cannot entirely explain - it *also* works if I
just add "VERSION { { global: *; }; }" instead, indicating that
something else is happening that I don't really understand. It
may be that explicitly doing that marks them with some kind of
empty version, and that's different from the default.

Explicitly marking them with a version breaks kallsyms, so that
doesn't seem to be possible.

Marking all the symbols as local seems correct, and does seem
to address the issue, so do that. Also do it for static link,
nsswitch libraries could still be loaded there.

[1] https://bugs.debian.org/983379

Reported-by: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Tested-By: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue May 23, 2021
[ Upstream commit d5027ca ]

Ritesh reported a bug [1] against UML, noting that it crashed on
startup. The backtrace shows the following (heavily redacted):

(gdb) bt
...
 #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268
 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2
 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72
...
 #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359
...
 #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486
 #45 0x00007f8990968b85 in __getgrnam_r [...]
 #46 0x00007f89909d6b77 in grantpt [...]
 #47 0x00007f8990a9394e in __GI_openpty [...]
 #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407
 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598
 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45
 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334
 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144

indicating that the UML function openpty_cb() calls openpty(),
which internally calls __getgrnam_r(), which causes the nsswitch
machinery to get started.

This loads, through lots of indirection that I snipped, the
libcom_err.so.2 library, which (in an unknown function, "??")
calls sem_init().

Now, of course it wants to get libpthread's sem_init(), since
it's linked against libpthread. However, the dynamic linker
looks up that symbol against the binary first, and gets the
kernel's sem_init().

Hajime Tazaki noted that "objcopy -L" can localize a symbol,
so the dynamic linker wouldn't do the lookup this way. I tried,
but for some reason that didn't seem to work.

Doing the same thing in the linker script instead does seem to
work, though I cannot entirely explain - it *also* works if I
just add "VERSION { { global: *; }; }" instead, indicating that
something else is happening that I don't really understand. It
may be that explicitly doing that marks them with some kind of
empty version, and that's different from the default.

Explicitly marking them with a version breaks kallsyms, so that
doesn't seem to be possible.

Marking all the symbols as local seems correct, and does seem
to address the issue, so do that. Also do it for static link,
nsswitch libraries could still be loaded there.

[1] https://bugs.debian.org/983379

Reported-by: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Tested-By: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue May 25, 2021
[ Upstream commit d5027ca ]

Ritesh reported a bug [1] against UML, noting that it crashed on
startup. The backtrace shows the following (heavily redacted):

(gdb) bt
...
 #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268
 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2
 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72
...
 #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359
...
 #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486
 #45 0x00007f8990968b85 in __getgrnam_r [...]
 #46 0x00007f89909d6b77 in grantpt [...]
 #47 0x00007f8990a9394e in __GI_openpty [...]
 #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407
 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598
 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45
 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334
 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144

indicating that the UML function openpty_cb() calls openpty(),
which internally calls __getgrnam_r(), which causes the nsswitch
machinery to get started.

This loads, through lots of indirection that I snipped, the
libcom_err.so.2 library, which (in an unknown function, "??")
calls sem_init().

Now, of course it wants to get libpthread's sem_init(), since
it's linked against libpthread. However, the dynamic linker
looks up that symbol against the binary first, and gets the
kernel's sem_init().

Hajime Tazaki noted that "objcopy -L" can localize a symbol,
so the dynamic linker wouldn't do the lookup this way. I tried,
but for some reason that didn't seem to work.

Doing the same thing in the linker script instead does seem to
work, though I cannot entirely explain - it *also* works if I
just add "VERSION { { global: *; }; }" instead, indicating that
something else is happening that I don't really understand. It
may be that explicitly doing that marks them with some kind of
empty version, and that's different from the default.

Explicitly marking them with a version breaks kallsyms, so that
doesn't seem to be possible.

Marking all the symbols as local seems correct, and does seem
to address the issue, so do that. Also do it for static link,
nsswitch libraries could still be loaded there.

[1] https://bugs.debian.org/983379

Reported-by: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Tested-By: Ritesh Raj Sarraf <rrs@debian.org>
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue Jan 18, 2023
[ Upstream commit 031af50 ]

The inline assembly for arm64's cmpxchg_double*() implementations use a
+Q constraint to hazard against other accesses to the memory location
being exchanged. However, the pointer passed to the constraint is a
pointer to unsigned long, and thus the hazard only applies to the first
8 bytes of the location.

GCC can take advantage of this, assuming that other portions of the
location are unchanged, leading to a number of potential problems.

This is similar to what we fixed back in commit:

  fee960b ("arm64: xchg: hazard against entire exchange variable")

... but we forgot to adjust cmpxchg_double*() similarly at the same
time.

The same problem applies, as demonstrated with the following test:

| struct big {
|         u64 lo, hi;
| } __aligned(128);
|
| unsigned long foo(struct big *b)
| {
|         u64 hi_old, hi_new;
|
|         hi_old = b->hi;
|         cmpxchg_double_local(&b->lo, &b->hi, 0x12, 0x34, 0x56, 0x78);
|         hi_new = b->hi;
|
|         return hi_old ^ hi_new;
| }

... which GCC 12.1.0 compiles as:

| 0000000000000000 <foo>:
|    0:   d503233f        paciasp
|    4:   aa0003e4        mov     x4, x0
|    8:   1400000e        b       40 <foo+0x40>
|    c:   d2800240        mov     x0, #0x12                       // #18
|   10:   d2800681        mov     x1, #0x34                       // #52
|   14:   aa0003e5        mov     x5, x0
|   18:   aa0103e6        mov     x6, x1
|   1c:   d2800ac2        mov     x2, #0x56                       // #86
|   20:   d2800f03        mov     x3, #0x78                       // #120
|   24:   48207c82        casp    x0, x1, x2, x3, [x4]
|   28:   ca050000        eor     x0, x0, x5
|   2c:   ca060021        eor     x1, x1, x6
|   30:   aa010000        orr     x0, x0, x1
|   34:   d2800000        mov     x0, #0x0                        // #0    <--- BANG
|   38:   d50323bf        autiasp
|   3c:   d65f03c0        ret
|   40:   d2800240        mov     x0, #0x12                       // #18
|   44:   d2800681        mov     x1, #0x34                       // #52
|   48:   d2800ac2        mov     x2, #0x56                       // #86
|   4c:   d2800f03        mov     x3, #0x78                       // #120
|   50:   f9800091        prfm    pstl1strm, [x4]
|   54:   c87f1885        ldxp    x5, x6, [x4]
|   58:   ca0000a5        eor     x5, x5, x0
|   5c:   ca0100c6        eor     x6, x6, x1
|   60:   aa0600a6        orr     x6, x5, x6
|   64:   b5000066        cbnz    x6, 70 <foo+0x70>
|   68:   c8250c82        stxp    w5, x2, x3, [x4]
|   6c:   35ffff45        cbnz    w5, 54 <foo+0x54>
|   70:   d2800000        mov     x0, #0x0                        // #0     <--- BANG
|   74:   d50323bf        autiasp
|   78:   d65f03c0        ret

Notice that at the lines with "BANG" comments, GCC has assumed that the
higher 8 bytes are unchanged by the cmpxchg_double() call, and that
`hi_old ^ hi_new` can be reduced to a constant zero, for both LSE and
LL/SC versions of cmpxchg_double().

This patch fixes the issue by passing a pointer to __uint128_t into the
+Q constraint, ensuring that the compiler hazards against the entire 16
bytes being modified.

With this change, GCC 12.1.0 compiles the above test as:

| 0000000000000000 <foo>:
|    0:   f9400407        ldr     x7, [x0, #8]
|    4:   d503233f        paciasp
|    8:   aa0003e4        mov     x4, x0
|    c:   1400000f        b       48 <foo+0x48>
|   10:   d2800240        mov     x0, #0x12                       // #18
|   14:   d2800681        mov     x1, #0x34                       // #52
|   18:   aa0003e5        mov     x5, x0
|   1c:   aa0103e6        mov     x6, x1
|   20:   d2800ac2        mov     x2, #0x56                       // #86
|   24:   d2800f03        mov     x3, #0x78                       // #120
|   28:   48207c82        casp    x0, x1, x2, x3, [x4]
|   2c:   ca050000        eor     x0, x0, x5
|   30:   ca060021        eor     x1, x1, x6
|   34:   aa010000        orr     x0, x0, x1
|   38:   f9400480        ldr     x0, [x4, #8]
|   3c:   d50323bf        autiasp
|   40:   ca0000e0        eor     x0, x7, x0
|   44:   d65f03c0        ret
|   48:   d2800240        mov     x0, #0x12                       // #18
|   4c:   d2800681        mov     x1, #0x34                       // #52
|   50:   d2800ac2        mov     x2, #0x56                       // #86
|   54:   d2800f03        mov     x3, #0x78                       // #120
|   58:   f9800091        prfm    pstl1strm, [x4]
|   5c:   c87f1885        ldxp    x5, x6, [x4]
|   60:   ca0000a5        eor     x5, x5, x0
|   64:   ca0100c6        eor     x6, x6, x1
|   68:   aa0600a6        orr     x6, x5, x6
|   6c:   b5000066        cbnz    x6, 78 <foo+0x78>
|   70:   c8250c82        stxp    w5, x2, x3, [x4]
|   74:   35ffff45        cbnz    w5, 5c <foo+0x5c>
|   78:   f9400480        ldr     x0, [x4, #8]
|   7c:   d50323bf        autiasp
|   80:   ca0000e0        eor     x0, x7, x0
|   84:   d65f03c0        ret

... sampling the high 8 bytes before and after the cmpxchg, and
performing an EOR, as we'd expect.

For backporting, I've tested this atop linux-4.9.y with GCC 5.5.0. Note
that linux-4.9.y is oldest currently supported stable release, and
mandates GCC 5.1+. Unfortunately I couldn't get a GCC 5.1 binary to run
on my machines due to library incompatibilities.

I've also used a standalone test to check that we can use a __uint128_t
pointer in a +Q constraint at least as far back as GCC 4.8.5 and LLVM
3.9.1.

Fixes: 5284e1b ("arm64: xchg: Implement cmpxchg_double")
Fixes: e9a4b79 ("arm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU")
Reported-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/lkml/Y6DEfQXymYVgL3oJ@boqun-archlinux/
Reported-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/lkml/Y6GXoO4qmH9OIZ5Q@hirez.programming.kicks-ass.net/
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: stable@vger.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20230104151626.3262137-1-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue Jan 18, 2023
[ Upstream commit 031af50 ]

The inline assembly for arm64's cmpxchg_double*() implementations use a
+Q constraint to hazard against other accesses to the memory location
being exchanged. However, the pointer passed to the constraint is a
pointer to unsigned long, and thus the hazard only applies to the first
8 bytes of the location.

GCC can take advantage of this, assuming that other portions of the
location are unchanged, leading to a number of potential problems.

This is similar to what we fixed back in commit:

  fee960b ("arm64: xchg: hazard against entire exchange variable")

... but we forgot to adjust cmpxchg_double*() similarly at the same
time.

The same problem applies, as demonstrated with the following test:

| struct big {
|         u64 lo, hi;
| } __aligned(128);
|
| unsigned long foo(struct big *b)
| {
|         u64 hi_old, hi_new;
|
|         hi_old = b->hi;
|         cmpxchg_double_local(&b->lo, &b->hi, 0x12, 0x34, 0x56, 0x78);
|         hi_new = b->hi;
|
|         return hi_old ^ hi_new;
| }

... which GCC 12.1.0 compiles as:

| 0000000000000000 <foo>:
|    0:   d503233f        paciasp
|    4:   aa0003e4        mov     x4, x0
|    8:   1400000e        b       40 <foo+0x40>
|    c:   d2800240        mov     x0, #0x12                       // #18
|   10:   d2800681        mov     x1, #0x34                       // #52
|   14:   aa0003e5        mov     x5, x0
|   18:   aa0103e6        mov     x6, x1
|   1c:   d2800ac2        mov     x2, #0x56                       // #86
|   20:   d2800f03        mov     x3, #0x78                       // #120
|   24:   48207c82        casp    x0, x1, x2, x3, [x4]
|   28:   ca050000        eor     x0, x0, x5
|   2c:   ca060021        eor     x1, x1, x6
|   30:   aa010000        orr     x0, x0, x1
|   34:   d2800000        mov     x0, #0x0                        // #0    <--- BANG
|   38:   d50323bf        autiasp
|   3c:   d65f03c0        ret
|   40:   d2800240        mov     x0, #0x12                       // #18
|   44:   d2800681        mov     x1, #0x34                       // #52
|   48:   d2800ac2        mov     x2, #0x56                       // #86
|   4c:   d2800f03        mov     x3, #0x78                       // #120
|   50:   f9800091        prfm    pstl1strm, [x4]
|   54:   c87f1885        ldxp    x5, x6, [x4]
|   58:   ca0000a5        eor     x5, x5, x0
|   5c:   ca0100c6        eor     x6, x6, x1
|   60:   aa0600a6        orr     x6, x5, x6
|   64:   b5000066        cbnz    x6, 70 <foo+0x70>
|   68:   c8250c82        stxp    w5, x2, x3, [x4]
|   6c:   35ffff45        cbnz    w5, 54 <foo+0x54>
|   70:   d2800000        mov     x0, #0x0                        // #0     <--- BANG
|   74:   d50323bf        autiasp
|   78:   d65f03c0        ret

Notice that at the lines with "BANG" comments, GCC has assumed that the
higher 8 bytes are unchanged by the cmpxchg_double() call, and that
`hi_old ^ hi_new` can be reduced to a constant zero, for both LSE and
LL/SC versions of cmpxchg_double().

This patch fixes the issue by passing a pointer to __uint128_t into the
+Q constraint, ensuring that the compiler hazards against the entire 16
bytes being modified.

With this change, GCC 12.1.0 compiles the above test as:

| 0000000000000000 <foo>:
|    0:   f9400407        ldr     x7, [x0, #8]
|    4:   d503233f        paciasp
|    8:   aa0003e4        mov     x4, x0
|    c:   1400000f        b       48 <foo+0x48>
|   10:   d2800240        mov     x0, #0x12                       // #18
|   14:   d2800681        mov     x1, #0x34                       // #52
|   18:   aa0003e5        mov     x5, x0
|   1c:   aa0103e6        mov     x6, x1
|   20:   d2800ac2        mov     x2, #0x56                       // #86
|   24:   d2800f03        mov     x3, #0x78                       // #120
|   28:   48207c82        casp    x0, x1, x2, x3, [x4]
|   2c:   ca050000        eor     x0, x0, x5
|   30:   ca060021        eor     x1, x1, x6
|   34:   aa010000        orr     x0, x0, x1
|   38:   f9400480        ldr     x0, [x4, #8]
|   3c:   d50323bf        autiasp
|   40:   ca0000e0        eor     x0, x7, x0
|   44:   d65f03c0        ret
|   48:   d2800240        mov     x0, #0x12                       // #18
|   4c:   d2800681        mov     x1, #0x34                       // #52
|   50:   d2800ac2        mov     x2, #0x56                       // #86
|   54:   d2800f03        mov     x3, #0x78                       // #120
|   58:   f9800091        prfm    pstl1strm, [x4]
|   5c:   c87f1885        ldxp    x5, x6, [x4]
|   60:   ca0000a5        eor     x5, x5, x0
|   64:   ca0100c6        eor     x6, x6, x1
|   68:   aa0600a6        orr     x6, x5, x6
|   6c:   b5000066        cbnz    x6, 78 <foo+0x78>
|   70:   c8250c82        stxp    w5, x2, x3, [x4]
|   74:   35ffff45        cbnz    w5, 5c <foo+0x5c>
|   78:   f9400480        ldr     x0, [x4, #8]
|   7c:   d50323bf        autiasp
|   80:   ca0000e0        eor     x0, x7, x0
|   84:   d65f03c0        ret

... sampling the high 8 bytes before and after the cmpxchg, and
performing an EOR, as we'd expect.

For backporting, I've tested this atop linux-4.9.y with GCC 5.5.0. Note
that linux-4.9.y is oldest currently supported stable release, and
mandates GCC 5.1+. Unfortunately I couldn't get a GCC 5.1 binary to run
on my machines due to library incompatibilities.

I've also used a standalone test to check that we can use a __uint128_t
pointer in a +Q constraint at least as far back as GCC 4.8.5 and LLVM
3.9.1.

Fixes: 5284e1b ("arm64: xchg: Implement cmpxchg_double")
Fixes: e9a4b79 ("arm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU")
Reported-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/lkml/Y6DEfQXymYVgL3oJ@boqun-archlinux/
Reported-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/lkml/Y6GXoO4qmH9OIZ5Q@hirez.programming.kicks-ass.net/
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: stable@vger.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20230104151626.3262137-1-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue Jan 18, 2023
[ Upstream commit 031af50 ]

The inline assembly for arm64's cmpxchg_double*() implementations use a
+Q constraint to hazard against other accesses to the memory location
being exchanged. However, the pointer passed to the constraint is a
pointer to unsigned long, and thus the hazard only applies to the first
8 bytes of the location.

GCC can take advantage of this, assuming that other portions of the
location are unchanged, leading to a number of potential problems.

This is similar to what we fixed back in commit:

  fee960b ("arm64: xchg: hazard against entire exchange variable")

... but we forgot to adjust cmpxchg_double*() similarly at the same
time.

The same problem applies, as demonstrated with the following test:

| struct big {
|         u64 lo, hi;
| } __aligned(128);
|
| unsigned long foo(struct big *b)
| {
|         u64 hi_old, hi_new;
|
|         hi_old = b->hi;
|         cmpxchg_double_local(&b->lo, &b->hi, 0x12, 0x34, 0x56, 0x78);
|         hi_new = b->hi;
|
|         return hi_old ^ hi_new;
| }

... which GCC 12.1.0 compiles as:

| 0000000000000000 <foo>:
|    0:   d503233f        paciasp
|    4:   aa0003e4        mov     x4, x0
|    8:   1400000e        b       40 <foo+0x40>
|    c:   d2800240        mov     x0, #0x12                       // #18
|   10:   d2800681        mov     x1, #0x34                       // #52
|   14:   aa0003e5        mov     x5, x0
|   18:   aa0103e6        mov     x6, x1
|   1c:   d2800ac2        mov     x2, #0x56                       // #86
|   20:   d2800f03        mov     x3, #0x78                       // #120
|   24:   48207c82        casp    x0, x1, x2, x3, [x4]
|   28:   ca050000        eor     x0, x0, x5
|   2c:   ca060021        eor     x1, x1, x6
|   30:   aa010000        orr     x0, x0, x1
|   34:   d2800000        mov     x0, #0x0                        // #0    <--- BANG
|   38:   d50323bf        autiasp
|   3c:   d65f03c0        ret
|   40:   d2800240        mov     x0, #0x12                       // #18
|   44:   d2800681        mov     x1, #0x34                       // #52
|   48:   d2800ac2        mov     x2, #0x56                       // #86
|   4c:   d2800f03        mov     x3, #0x78                       // #120
|   50:   f9800091        prfm    pstl1strm, [x4]
|   54:   c87f1885        ldxp    x5, x6, [x4]
|   58:   ca0000a5        eor     x5, x5, x0
|   5c:   ca0100c6        eor     x6, x6, x1
|   60:   aa0600a6        orr     x6, x5, x6
|   64:   b5000066        cbnz    x6, 70 <foo+0x70>
|   68:   c8250c82        stxp    w5, x2, x3, [x4]
|   6c:   35ffff45        cbnz    w5, 54 <foo+0x54>
|   70:   d2800000        mov     x0, #0x0                        // #0     <--- BANG
|   74:   d50323bf        autiasp
|   78:   d65f03c0        ret

Notice that at the lines with "BANG" comments, GCC has assumed that the
higher 8 bytes are unchanged by the cmpxchg_double() call, and that
`hi_old ^ hi_new` can be reduced to a constant zero, for both LSE and
LL/SC versions of cmpxchg_double().

This patch fixes the issue by passing a pointer to __uint128_t into the
+Q constraint, ensuring that the compiler hazards against the entire 16
bytes being modified.

With this change, GCC 12.1.0 compiles the above test as:

| 0000000000000000 <foo>:
|    0:   f9400407        ldr     x7, [x0, #8]
|    4:   d503233f        paciasp
|    8:   aa0003e4        mov     x4, x0
|    c:   1400000f        b       48 <foo+0x48>
|   10:   d2800240        mov     x0, #0x12                       // #18
|   14:   d2800681        mov     x1, #0x34                       // #52
|   18:   aa0003e5        mov     x5, x0
|   1c:   aa0103e6        mov     x6, x1
|   20:   d2800ac2        mov     x2, #0x56                       // #86
|   24:   d2800f03        mov     x3, #0x78                       // #120
|   28:   48207c82        casp    x0, x1, x2, x3, [x4]
|   2c:   ca050000        eor     x0, x0, x5
|   30:   ca060021        eor     x1, x1, x6
|   34:   aa010000        orr     x0, x0, x1
|   38:   f9400480        ldr     x0, [x4, #8]
|   3c:   d50323bf        autiasp
|   40:   ca0000e0        eor     x0, x7, x0
|   44:   d65f03c0        ret
|   48:   d2800240        mov     x0, #0x12                       // #18
|   4c:   d2800681        mov     x1, #0x34                       // #52
|   50:   d2800ac2        mov     x2, #0x56                       // #86
|   54:   d2800f03        mov     x3, #0x78                       // #120
|   58:   f9800091        prfm    pstl1strm, [x4]
|   5c:   c87f1885        ldxp    x5, x6, [x4]
|   60:   ca0000a5        eor     x5, x5, x0
|   64:   ca0100c6        eor     x6, x6, x1
|   68:   aa0600a6        orr     x6, x5, x6
|   6c:   b5000066        cbnz    x6, 78 <foo+0x78>
|   70:   c8250c82        stxp    w5, x2, x3, [x4]
|   74:   35ffff45        cbnz    w5, 5c <foo+0x5c>
|   78:   f9400480        ldr     x0, [x4, #8]
|   7c:   d50323bf        autiasp
|   80:   ca0000e0        eor     x0, x7, x0
|   84:   d65f03c0        ret

... sampling the high 8 bytes before and after the cmpxchg, and
performing an EOR, as we'd expect.

For backporting, I've tested this atop linux-4.9.y with GCC 5.5.0. Note
that linux-4.9.y is oldest currently supported stable release, and
mandates GCC 5.1+. Unfortunately I couldn't get a GCC 5.1 binary to run
on my machines due to library incompatibilities.

I've also used a standalone test to check that we can use a __uint128_t
pointer in a +Q constraint at least as far back as GCC 4.8.5 and LLVM
3.9.1.

Fixes: 5284e1b ("arm64: xchg: Implement cmpxchg_double")
Fixes: e9a4b79 ("arm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU")
Reported-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/lkml/Y6DEfQXymYVgL3oJ@boqun-archlinux/
Reported-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/lkml/Y6GXoO4qmH9OIZ5Q@hirez.programming.kicks-ass.net/
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: stable@vger.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20230104151626.3262137-1-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue Jan 18, 2023
[ Upstream commit 031af50 ]

The inline assembly for arm64's cmpxchg_double*() implementations use a
+Q constraint to hazard against other accesses to the memory location
being exchanged. However, the pointer passed to the constraint is a
pointer to unsigned long, and thus the hazard only applies to the first
8 bytes of the location.

GCC can take advantage of this, assuming that other portions of the
location are unchanged, leading to a number of potential problems.

This is similar to what we fixed back in commit:

  fee960b ("arm64: xchg: hazard against entire exchange variable")

... but we forgot to adjust cmpxchg_double*() similarly at the same
time.

The same problem applies, as demonstrated with the following test:

| struct big {
|         u64 lo, hi;
| } __aligned(128);
|
| unsigned long foo(struct big *b)
| {
|         u64 hi_old, hi_new;
|
|         hi_old = b->hi;
|         cmpxchg_double_local(&b->lo, &b->hi, 0x12, 0x34, 0x56, 0x78);
|         hi_new = b->hi;
|
|         return hi_old ^ hi_new;
| }

... which GCC 12.1.0 compiles as:

| 0000000000000000 <foo>:
|    0:   d503233f        paciasp
|    4:   aa0003e4        mov     x4, x0
|    8:   1400000e        b       40 <foo+0x40>
|    c:   d2800240        mov     x0, #0x12                       // #18
|   10:   d2800681        mov     x1, #0x34                       // #52
|   14:   aa0003e5        mov     x5, x0
|   18:   aa0103e6        mov     x6, x1
|   1c:   d2800ac2        mov     x2, #0x56                       // #86
|   20:   d2800f03        mov     x3, #0x78                       // #120
|   24:   48207c82        casp    x0, x1, x2, x3, [x4]
|   28:   ca050000        eor     x0, x0, x5
|   2c:   ca060021        eor     x1, x1, x6
|   30:   aa010000        orr     x0, x0, x1
|   34:   d2800000        mov     x0, #0x0                        // #0    <--- BANG
|   38:   d50323bf        autiasp
|   3c:   d65f03c0        ret
|   40:   d2800240        mov     x0, #0x12                       // #18
|   44:   d2800681        mov     x1, #0x34                       // #52
|   48:   d2800ac2        mov     x2, #0x56                       // #86
|   4c:   d2800f03        mov     x3, #0x78                       // #120
|   50:   f9800091        prfm    pstl1strm, [x4]
|   54:   c87f1885        ldxp    x5, x6, [x4]
|   58:   ca0000a5        eor     x5, x5, x0
|   5c:   ca0100c6        eor     x6, x6, x1
|   60:   aa0600a6        orr     x6, x5, x6
|   64:   b5000066        cbnz    x6, 70 <foo+0x70>
|   68:   c8250c82        stxp    w5, x2, x3, [x4]
|   6c:   35ffff45        cbnz    w5, 54 <foo+0x54>
|   70:   d2800000        mov     x0, #0x0                        // #0     <--- BANG
|   74:   d50323bf        autiasp
|   78:   d65f03c0        ret

Notice that at the lines with "BANG" comments, GCC has assumed that the
higher 8 bytes are unchanged by the cmpxchg_double() call, and that
`hi_old ^ hi_new` can be reduced to a constant zero, for both LSE and
LL/SC versions of cmpxchg_double().

This patch fixes the issue by passing a pointer to __uint128_t into the
+Q constraint, ensuring that the compiler hazards against the entire 16
bytes being modified.

With this change, GCC 12.1.0 compiles the above test as:

| 0000000000000000 <foo>:
|    0:   f9400407        ldr     x7, [x0, #8]
|    4:   d503233f        paciasp
|    8:   aa0003e4        mov     x4, x0
|    c:   1400000f        b       48 <foo+0x48>
|   10:   d2800240        mov     x0, #0x12                       // #18
|   14:   d2800681        mov     x1, #0x34                       // #52
|   18:   aa0003e5        mov     x5, x0
|   1c:   aa0103e6        mov     x6, x1
|   20:   d2800ac2        mov     x2, #0x56                       // #86
|   24:   d2800f03        mov     x3, #0x78                       // #120
|   28:   48207c82        casp    x0, x1, x2, x3, [x4]
|   2c:   ca050000        eor     x0, x0, x5
|   30:   ca060021        eor     x1, x1, x6
|   34:   aa010000        orr     x0, x0, x1
|   38:   f9400480        ldr     x0, [x4, #8]
|   3c:   d50323bf        autiasp
|   40:   ca0000e0        eor     x0, x7, x0
|   44:   d65f03c0        ret
|   48:   d2800240        mov     x0, #0x12                       // #18
|   4c:   d2800681        mov     x1, #0x34                       // #52
|   50:   d2800ac2        mov     x2, #0x56                       // #86
|   54:   d2800f03        mov     x3, #0x78                       // #120
|   58:   f9800091        prfm    pstl1strm, [x4]
|   5c:   c87f1885        ldxp    x5, x6, [x4]
|   60:   ca0000a5        eor     x5, x5, x0
|   64:   ca0100c6        eor     x6, x6, x1
|   68:   aa0600a6        orr     x6, x5, x6
|   6c:   b5000066        cbnz    x6, 78 <foo+0x78>
|   70:   c8250c82        stxp    w5, x2, x3, [x4]
|   74:   35ffff45        cbnz    w5, 5c <foo+0x5c>
|   78:   f9400480        ldr     x0, [x4, #8]
|   7c:   d50323bf        autiasp
|   80:   ca0000e0        eor     x0, x7, x0
|   84:   d65f03c0        ret

... sampling the high 8 bytes before and after the cmpxchg, and
performing an EOR, as we'd expect.

For backporting, I've tested this atop linux-4.9.y with GCC 5.5.0. Note
that linux-4.9.y is oldest currently supported stable release, and
mandates GCC 5.1+. Unfortunately I couldn't get a GCC 5.1 binary to run
on my machines due to library incompatibilities.

I've also used a standalone test to check that we can use a __uint128_t
pointer in a +Q constraint at least as far back as GCC 4.8.5 and LLVM
3.9.1.

Fixes: 5284e1b ("arm64: xchg: Implement cmpxchg_double")
Fixes: e9a4b79 ("arm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU")
Reported-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/lkml/Y6DEfQXymYVgL3oJ@boqun-archlinux/
Reported-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/lkml/Y6GXoO4qmH9OIZ5Q@hirez.programming.kicks-ass.net/
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: stable@vger.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20230104151626.3262137-1-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
anthraxx pushed a commit that referenced this issue Jan 18, 2023
commit 031af50 upstream.

The inline assembly for arm64's cmpxchg_double*() implementations use a
+Q constraint to hazard against other accesses to the memory location
being exchanged. However, the pointer passed to the constraint is a
pointer to unsigned long, and thus the hazard only applies to the first
8 bytes of the location.

GCC can take advantage of this, assuming that other portions of the
location are unchanged, leading to a number of potential problems.

This is similar to what we fixed back in commit:

  fee960b ("arm64: xchg: hazard against entire exchange variable")

... but we forgot to adjust cmpxchg_double*() similarly at the same
time.

The same problem applies, as demonstrated with the following test:

| struct big {
|         u64 lo, hi;
| } __aligned(128);
|
| unsigned long foo(struct big *b)
| {
|         u64 hi_old, hi_new;
|
|         hi_old = b->hi;
|         cmpxchg_double_local(&b->lo, &b->hi, 0x12, 0x34, 0x56, 0x78);
|         hi_new = b->hi;
|
|         return hi_old ^ hi_new;
| }

... which GCC 12.1.0 compiles as:

| 0000000000000000 <foo>:
|    0:   d503233f        paciasp
|    4:   aa0003e4        mov     x4, x0
|    8:   1400000e        b       40 <foo+0x40>
|    c:   d2800240        mov     x0, #0x12                       // #18
|   10:   d2800681        mov     x1, #0x34                       // #52
|   14:   aa0003e5        mov     x5, x0
|   18:   aa0103e6        mov     x6, x1
|   1c:   d2800ac2        mov     x2, #0x56                       // #86
|   20:   d2800f03        mov     x3, #0x78                       // #120
|   24:   48207c82        casp    x0, x1, x2, x3, [x4]
|   28:   ca050000        eor     x0, x0, x5
|   2c:   ca060021        eor     x1, x1, x6
|   30:   aa010000        orr     x0, x0, x1
|   34:   d2800000        mov     x0, #0x0                        // #0    <--- BANG
|   38:   d50323bf        autiasp
|   3c:   d65f03c0        ret
|   40:   d2800240        mov     x0, #0x12                       // #18
|   44:   d2800681        mov     x1, #0x34                       // #52
|   48:   d2800ac2        mov     x2, #0x56                       // #86
|   4c:   d2800f03        mov     x3, #0x78                       // #120
|   50:   f9800091        prfm    pstl1strm, [x4]
|   54:   c87f1885        ldxp    x5, x6, [x4]
|   58:   ca0000a5        eor     x5, x5, x0
|   5c:   ca0100c6        eor     x6, x6, x1
|   60:   aa0600a6        orr     x6, x5, x6
|   64:   b5000066        cbnz    x6, 70 <foo+0x70>
|   68:   c8250c82        stxp    w5, x2, x3, [x4]
|   6c:   35ffff45        cbnz    w5, 54 <foo+0x54>
|   70:   d2800000        mov     x0, #0x0                        // #0     <--- BANG
|   74:   d50323bf        autiasp
|   78:   d65f03c0        ret

Notice that at the lines with "BANG" comments, GCC has assumed that the
higher 8 bytes are unchanged by the cmpxchg_double() call, and that
`hi_old ^ hi_new` can be reduced to a constant zero, for both LSE and
LL/SC versions of cmpxchg_double().

This patch fixes the issue by passing a pointer to __uint128_t into the
+Q constraint, ensuring that the compiler hazards against the entire 16
bytes being modified.

With this change, GCC 12.1.0 compiles the above test as:

| 0000000000000000 <foo>:
|    0:   f9400407        ldr     x7, [x0, #8]
|    4:   d503233f        paciasp
|    8:   aa0003e4        mov     x4, x0
|    c:   1400000f        b       48 <foo+0x48>
|   10:   d2800240        mov     x0, #0x12                       // #18
|   14:   d2800681        mov     x1, #0x34                       // #52
|   18:   aa0003e5        mov     x5, x0
|   1c:   aa0103e6        mov     x6, x1
|   20:   d2800ac2        mov     x2, #0x56                       // #86
|   24:   d2800f03        mov     x3, #0x78                       // #120
|   28:   48207c82        casp    x0, x1, x2, x3, [x4]
|   2c:   ca050000        eor     x0, x0, x5
|   30:   ca060021        eor     x1, x1, x6
|   34:   aa010000        orr     x0, x0, x1
|   38:   f9400480        ldr     x0, [x4, #8]
|   3c:   d50323bf        autiasp
|   40:   ca0000e0        eor     x0, x7, x0
|   44:   d65f03c0        ret
|   48:   d2800240        mov     x0, #0x12                       // #18
|   4c:   d2800681        mov     x1, #0x34                       // #52
|   50:   d2800ac2        mov     x2, #0x56                       // #86
|   54:   d2800f03        mov     x3, #0x78                       // #120
|   58:   f9800091        prfm    pstl1strm, [x4]
|   5c:   c87f1885        ldxp    x5, x6, [x4]
|   60:   ca0000a5        eor     x5, x5, x0
|   64:   ca0100c6        eor     x6, x6, x1
|   68:   aa0600a6        orr     x6, x5, x6
|   6c:   b5000066        cbnz    x6, 78 <foo+0x78>
|   70:   c8250c82        stxp    w5, x2, x3, [x4]
|   74:   35ffff45        cbnz    w5, 5c <foo+0x5c>
|   78:   f9400480        ldr     x0, [x4, #8]
|   7c:   d50323bf        autiasp
|   80:   ca0000e0        eor     x0, x7, x0
|   84:   d65f03c0        ret

... sampling the high 8 bytes before and after the cmpxchg, and
performing an EOR, as we'd expect.

For backporting, I've tested this atop linux-4.9.y with GCC 5.5.0. Note
that linux-4.9.y is oldest currently supported stable release, and
mandates GCC 5.1+. Unfortunately I couldn't get a GCC 5.1 binary to run
on my machines due to library incompatibilities.

I've also used a standalone test to check that we can use a __uint128_t
pointer in a +Q constraint at least as far back as GCC 4.8.5 and LLVM
3.9.1.

Fixes: 5284e1b ("arm64: xchg: Implement cmpxchg_double")
Fixes: e9a4b79 ("arm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU")
Reported-by: Boqun Feng <boqun.feng@gmail.com>
Link: https://lore.kernel.org/lkml/Y6DEfQXymYVgL3oJ@boqun-archlinux/
Reported-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/lkml/Y6GXoO4qmH9OIZ5Q@hirez.programming.kicks-ass.net/
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: stable@vger.kernel.org
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Steve Capper <steve.capper@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20230104151626.3262137-1-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working good first issue Good for newcomers
Projects
None yet
Development

No branches or pull requests

2 participants