[release-branch.go1.17] runtime: drop SIGPROF while in ARM < 7 kernel helpers

On Linux ARMv6 and below runtime/internal/atomic.Cas calls into a kernel
cas helper at a fixed address. If a SIGPROF arrives while executing the
kernel helper, the sigprof lostAtomic logic will miss that we are
potentially in the spinlock critical section, which could cause
a deadlock when using atomics later in sigprof.

For #47505
Fixes #47688

Change-Id: If8ba0d0fc47e45d4e6c68eca98fac4c6ed4e43c1
Reviewed-on: https://go-review.googlesource.com/c/go/+/341889
Trust: Michael Pratt <mpratt@google.com>
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
(cherry picked from commit 20a620fd9f7bc35739c1af3602d53808d0430814)
Reviewed-on: https://go-review.googlesource.com/c/go/+/341890

diff --git a/src/runtime/proc.go b/src/runtime/proc.go
index 7bc2a92590755ee3b1c40255d00c760fa30e927b..a144b36ec80b75cea5a41c7dabadceeaa18e9fcb 100644 (file)
--- a/src/runtime/proc.go
+++ b/src/runtime/proc.go
@@ -4686,7 +4686,7 @@ func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
                return
        }
 
-       // On mips{,le}, 64bit atomics are emulated with spinlocks, in
+       // On mips{,le}/arm, 64bit atomics are emulated with spinlocks, in
        // runtime/internal/atomic. If SIGPROF arrives while the program is inside
        // the critical section, it creates a deadlock (when writing the sample).
        // As a workaround, create a counter of SIGPROFs while in critical section
@@ -4699,6 +4699,13 @@ func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
                                return
                        }
                }
+               if GOARCH == "arm" && goarm < 7 && GOOS == "linux" && pc&0xffff0000 == 0xffff0000 {
+                       // runtime/internal/atomic functions call into kernel
+                       // helpers on arm < 7. See
+                       // runtime/internal/atomic/sys_linux_arm.s.
+                       cpuprof.lostAtomic++
+                       return
+               }
        }
 
        // Profiling runs concurrently with GC, so it must not allocate.