[Qemu-devel] [PATCH for-2.0? 2/2] cpu-exec: Unlock tb_lock if we longjmp out of code generation

[Peter Maydell]

If the guest attempts to execute from unreadable memory, this will
cause us to longjmp back to the main loop from inside the
target frontend decoder. For linux-user mode, this means we will
still hold the tb_ctx.tb_lock, and will deadlock when we try to
start executing code again. Unlock the lock in the return-from-longjmp
code path to avoid this.

Signed-off-by: Peter Maydell <address@hidden>
---
 cpu-exec.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/cpu-exec.c b/cpu-exec.c
index 0914d3c..02168d9 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -227,6 +227,8 @@ int cpu_exec(CPUArchState *env)
     TranslationBlock *tb;
     uint8_t *tc_ptr;
     uintptr_t next_tb;
+    /* This must be volatile so it is not trashed by longjmp() */
+    volatile bool have_tb_lock = false;
 
     if (cpu->halted) {
         if (!cpu_has_work(cpu)) {
@@ -600,6 +602,7 @@ int cpu_exec(CPUArchState *env)
                     cpu_loop_exit(cpu);
                 }
                 spin_lock(&tcg_ctx.tb_ctx.tb_lock);
+                have_tb_lock = true;
                 tb = tb_find_fast(env);
                 /* Note: we do it here to avoid a gcc bug on Mac OS X when
                    doing it in tb_find_slow */
@@ -621,6 +624,7 @@ int cpu_exec(CPUArchState *env)
                     tb_add_jump((TranslationBlock *)(next_tb & ~TB_EXIT_MASK),
                                 next_tb & TB_EXIT_MASK, tb);
                 }
+                have_tb_lock = false;
                 spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
 
                 /* cpu_interrupt might be called while translating the
@@ -692,6 +696,9 @@ int cpu_exec(CPUArchState *env)
 #ifdef TARGET_I386
             x86_cpu = X86_CPU(cpu);
 #endif
+            if (have_tb_lock) {
+                spin_unlock(&tcg_ctx.tb_ctx.tb_lock);
+            }
         }
     } /* for(;;) */
 
-- 
1.9.0