page_check_range: don't bail out early after unprotecting page

When checking a page range, if we found that a page was
made read-only by QEMU because it contained translated code,
we were incorrectly returning immediately after unprotecting
that page, rather than continuing to check the entire range,
so we might fail to unprotect pages later in the range, or
might incorrectly return a "success" result even if later
pages were not writable.

In particular, this could cause segfaults in a case where
signals are delivered back to back on a target architecture
which uses trampoline code in the stack frame (as AArch64
currently does). The second signal causes a segfault because
the frame cannot be written to (it was protected because
we translated and executed the restorer trampoline, and the
unprotect logic did not unprotect the whole range).

Signed-off-by: Andrei Warkentin <andrey.warkentin@gmail.com
[PMM: expanded commit message a bit]
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

diff --git a/translate-all.c b/translate-all.c
index f243c100949f7d8a4c9a163434f48f2913ceb387..5759974d4224c3b0c7990ec0418d67fee1458b5d 100644 (file)
--- a/translate-all.c
+++ b/translate-all.c
@@ -1777,7 +1777,6 @@ int page_check_range(target_ulong start, target_ulong len, int flags)
                     return -1;
                 }
             }
-            return 0;
         }
     }
     return 0;