[PATCH v1] LoongArch: Fix __smp_mb__{before,after}_atomic()

When testing the BPF selftest "sudo ./test_progs -t timer_lockup", there
is a kernel lockup and panic:

  watchdog: BUG: soft lockup - CPU#1 stuck for 8s! [test_progs:39601]
  Kernel panic - not syncing: softlockup: hung tasks
  ...
  Call Trace:
  [<9000000000c6d2a4>] show_stack+0xf4/0x1c0
  [<9000000000c66ad4>] dump_stack_lvl+0x84/0xc8
  [<9000000000c40c10>] vpanic+0x278/0x4bc
  [<9000000000c40e98>] panic+0x44/0x48
  [<9000000000e8b000>] watchdog_timer_fn+0x500/0x520
  [<9000000000dfd9a4>] __hrtimer_run_queues+0xc4/0x530
  [<9000000000dffe90>] hrtimer_interrupt+0x140/0x320
  [<9000000000c701e4>] constant_timer_interrupt+0x34/0x50
  [<9000000000d956b8>] __handle_irq_event_percpu+0xe8/0x560
  [<9000000000d95b4c>] handle_irq_event_percpu+0x1c/0x80
  [<9000000000d9e4c4>] handle_percpu_irq+0x74/0xd0
  [<9000000000d945d4>] generic_handle_domain_irq+0x34/0x90
  [<9000000001c4ee84>] handle_cpu_irq+0x64/0xa0
  [<9000000002ac0fa8>] handle_loongarch_irq+0x28/0x40
  [<9000000002ac103c>] do_vint+0x7c/0xf0
  [<9000000002ad9e4c>] _raw_spin_unlock_irqrestore+0x8c/0xc0
  [<9000000000dfe9e0>] hrtimer_try_to_cancel.part.0+0x70/0x350
  [<9000000000dfed58>] hrtimer_cancel+0x38/0x80
  [<9000000000f6d944>] bpf_timer_cancel+0x94/0x1e0
  [<ffff80000200fad0>] bpf_prog_108ab87b32f22e44_timer_cb1+0xb0/0xfc
  [<9000000000f6b838>] bpf_timer_cb+0x98/0x170
  [<9000000000dfdaac>] __hrtimer_run_queues+0x1cc/0x530
  [<9000000000dfde94>] hrtimer_run_softirq+0x84/0xd0
  [<9000000000caa584>] handle_softirqs+0x154/0x7a0
  [<9000000000cab0a4>] do_softirq+0xb4/0x120
  [<9000000000cab27c>] __local_bh_enable_ip+0x16c/0x1c0
  [<900000000271d9e0>] bpf_test_run+0x1c0/0x5c0
  [<900000000271f548>] bpf_prog_test_run_skb+0x6e8/0xe20
  [<9000000000f39940>] __sys_bpf+0x1690/0x2c50
  [<9000000000f3af28>] sys_bpf+0x28/0x40
  [<9000000002ac2d68>] do_syscall+0x108/0x5e0
  [<9000000000c6a850>] handle_syscall+0xd0/0x170

In bpf_timer_cancel() of kernel/bpf/helpers.c, it explicitly notes that
"Need full barrier after relaxed atomic_inc" to expect a full hardware
barrier to ensure global visibility, but there is only an empty barrier
for smp_mb__after_atomic().

As LoongArch is a weakly-ordered architecture, without a data barrier,
the physical ordering of "store-before-load" cannot be guaranteed. As
a result, a subsequent read can bypass a prior relaxed atomic write
during lockless dependency checks. This allows both CPUs to experience
store-load reordering and simultaneously bypass the lockless deadlock
detection in the software, leading them to proceed into hrtimer_cancel()
and trigger a severe ABBA deadlock in the BPF core during the concurrent
test runs, resulting in a kernel panic.

Upgrade __smp_mb__after_atomic() to __smp_mb(), which generates the data
barrier instruction DBAR to provide full barrier after relaxed atomic_inc.
While at it, also upgrade __smp_mb__before_atomic() to __smp_mb().

With this patch, the lockless "store-before-load" ordering is enforced by
the DBAR instruction. The BPF timer_lockup selftest was stressed for 5000
consecutive loops on a physical LoongArch machine without encountering any
further lockups or warnings:

  for i in {1..5000}; do sudo ./test_progs -t timer_lockup; done

Reported-by: Vincent Li <[email protected]>
Closes: https://lore.kernel.org/loongarch/CAK3+h2xOSEZUHhou7N2cRL-aGrZCNSm45g+P7thObMe+fpgYCA@mail.gmail.com/
Fixes: 46859ac8af52 ("LoongArch: Add multi-processor (SMP) support")

Signed-off-by: Tiezhu Yang <[email protected]>
---
 arch/loongarch/include/asm/barrier.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/loongarch/include/asm/barrier.h b/arch/loongarch/include/asm/barrier.h
index 4b663f197706..adfe343dfa65 100644
--- a/arch/loongarch/include/asm/barrier.h
+++ b/arch/loongarch/include/asm/barrier.h
@@ -57,8 +57,8 @@
 #define __WEAK_LLSC_MB		"			\n"
 #endif
 
-#define __smp_mb__before_atomic()	barrier()
-#define __smp_mb__after_atomic()	barrier()
+#define __smp_mb__before_atomic()	__smp_mb()
+#define __smp_mb__after_atomic()	__smp_mb()
 
 /**
  * array_index_mask_nospec() - generate a ~0 mask when index < size, 0 otherwise
-- 
2.42.0