syscall: apparent deadlock in `TestExecPtrace`

#!watchflakes
post <- pkg == "syscall" && log ~ `test timed out` && log ~ `\s+TestExecPtrace \(\d+m`

Issue created automatically to collect these failures.

Example (log):

panic: test timed out after 15m0s
running tests:
	TestExecPtrace (14m59s)

goroutine 44 gp=0x400018c000 m=7 mp=0x4000080708 [running]:
panic({0x14fe40?, 0x4000184050?})
	/home/gopher/build/go/src/runtime/panic.go:779 +0x140 fp=0x4000051f00 sp=0x4000051e50 pc=0x472c0
testing.(*M).startAlarm.func1()
	/home/gopher/build/go/src/testing/testing.go:2366 +0x30c fp=0x4000051fd0 sp=0x4000051f00 pc=0xec68c
runtime.goexit({})
...

goroutine 20 gp=0x400018c340 m=nil [GC worker (idle), 14 minutes]:
runtime.gopark(0x400003a7b0?, 0x0?, 0x0?, 0x0?, 0x0?)
	/home/gopher/build/go/src/runtime/proc.go:402 +0xc8 fp=0x400003a730 sp=0x400003a710 pc=0x4aad8
runtime.gcBgMarkWorker()
	/home/gopher/build/go/src/runtime/mgc.go:1310 +0xd8 fp=0x400003a7d0 sp=0x400003a730 pc=0x2ba18
runtime.goexit({})
	/home/gopher/build/go/src/runtime/asm_arm64.s:1222 +0x4 fp=0x400003a7d0 sp=0x400003a7d0 pc=0x823a4
created by runtime.gcBgMarkStartWorkers in goroutine 13
	/home/gopher/build/go/src/runtime/mgc.go:1234 +0x28

— watchflakes