Re: [Qemu-devel] [PATCH v2] thread-win32: fix GetThreadContext() permanently fails

[Zavadovsky Yan]

On Fri, Jun 26, 2015 at 1:28 PM, Olga Krishtal <address@hidden> wrote:

On 25/06/15 15:17, Zavadovsky Yan wrote:

On Thu, Jun 25, 2015 at 12:11 PM, Olga Krishtal <address@hidden>
wrote:

On 24/06/15 15:25, Zavadovsky Yan wrote:

Calling SuspendThread() is not enough to suspend Win32 thread.
We need to call GetThreadContext() after SuspendThread()
to make sure that OS have really suspended target thread.
But GetThreadContext() needs for THREAD_GET_CONTEXT
access right on thread object.
More info about this technique can be found here:
http://blogs.msdn.com/b/oldnewthing/archive/2015/02/05/10591215.aspx

This patch adds THREAD_GET_CONTEXT to OpenThread() arguments
and change oddity 'while(GetThreadContext() == SUCCESS)' to
'if(GetThreadContext() == FAILED){exit(1);}'.
So this block of code will continue only after successful
grabbing of thread context(i.e. when thread is really suspended).
And halts otherwise with more verbose error message than previous.
Signed-off-by: Zavadovsky Yan <address@hidden>
---
  cpus.c                   | 14 ++++++++------
  util/qemu-thread-win32.c |  4 ++--
  2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/cpus.c b/cpus.c
index 4f0e54d..0df6a7d 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1089,8 +1089,8 @@ static void qemu_cpu_kick_thread(CPUState *cpu)
          CONTEXT tcgContext;
            if (SuspendThread(cpu->hThread) == (DWORD)-1) {
-            fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
-                    GetLastError());
+            fprintf(stderr, "qemu:%s: SuspendThread GetLastError:%lu\n",
+                    __func__, GetLastError());
              exit(1);
          }
  @@ -1098,15 +1098,17 @@ static void qemu_cpu_kick_thread(CPUState *cpu)
           * suspended until we can get the context.
           */
          tcgContext.ContextFlags = CONTEXT_CONTROL;
-        while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
-            continue;

I would like to ask you if you have faced this situation in reality?
I have some doubts about changing the while() loop. According to the
article -

SuspendThread is async operation. But GetContext is not async. And
GetContext forces this pending suspension to complete.

thread may mot be suspended immediately after code, due to the busy

scheduler.
If we do exit(1) just right after check of GetThreadContext(..) we can
miss the situation when
scheduler is too busy at the moment and just go down.  From this point of
view this while -
is busy loop and gives the scheduler the opportunity to do its job.

If GetThreadContext fails in first call it will fail in next consecutive
calls for this thread handle.

Afaik GetThreadContext can fail if the thread is in running state, it means that
thread has not been suspended yet, for example it is running on other cpu with hight irql.

Citation from http://blogs.msdn.com/b/oldnewthing/archive/2015/02/05/10591215.aspx :

>If you want to make sure the thread really is suspended, you need to perform a synchronous operation that is dependent on the fact that the thread is suspended.

>The traditional way of doing this is to call Get­Thread­Context, since this requires the kernel to read from the context of the suspended thread, which has as a prerequisite that the context be saved in the first place, which has as a prerequisite that the thread be suspended.

It's written by MS insider. Qemu maintainers treat him as reliable source - http://lists.nongnu.org/archive/html/qemu-devel/2015-06/msg05929.html

Also you can compile sample from this article and run it. You will get "Huh? The variable lValue was modified by a suspended thread?" many times.

But if you modify this sample in this manner:

while (1) { memset(&Context, 0, sizeof(Context)); Context.ContextFlags = CONTEXT_CONTROL; if (SuspendThread(thread) == (DWORD)-1) abort(); if (GetThreadContext(thread, &Context) == 0) { printf("GetThreadContext failed\n"); } if (InterlockedOr(&lValue, 0) != InterlockedOr(&lValue, 0)) { printf("Huh? The variable lValue was modified by a suspended thread?\n"); } ResumeThread(thread); }

Compile, run and you will see that "Huh? The variable lValue was modified by a suspended thread?" message will not shown anymore.

 

Instead of waiting we are going to go down. I agree that busy loop should be somehow changed,
but going down every time.

So, I am not sure about it.

+        if (GetThreadContext(cpu->hThread, &tcgContext) == 0) {
+            fprintf(stderr, "qemu:%s: GetThreadContext
GetLastError:%lu\n",
+                    __func__, GetLastError());
+            exit(1);
          }
            cpu_signal(0);
            if (ResumeThread(cpu->hThread) == (DWORD)-1) {
-            fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
-                    GetLastError());
+            fprintf(stderr, "qemu:%s: ResumeThread GetLastError:%lu\n",
+                    __func__, GetLastError());
              exit(1);
          }
      }
diff --git a/util/qemu-thread-win32.c b/util/qemu-thread-win32.c
index 406b52f..823eca1 100644
--- a/util/qemu-thread-win32.c
+++ b/util/qemu-thread-win32.c
@@ -406,8 +406,8 @@ HANDLE qemu_thread_get_handle(QemuThread *thread)
        EnterCriticalSection(&data->cs);
      if (!data->exited) {
-        handle = OpenThread(SYNCHRONIZE | THREAD_SUSPEND_RESUME, FALSE,
-                            thread->tid);

What was before the usage of this flag?

GetThreadContext returns zero, i.e. error. GetLastError says "5" i.e.
"access denied".

I mean the behavior? As I can see it worked even without this flag.

This code works because:
a)GetThreadContext works as Sleep(N) and scheduler have enough time to
finish async suspend.
or
b)GetThreadContext always forces suspension to complete before checking
handle's desired access and doing its own job.
or
c)something else, documentation on SuspendThread is weak.

 +        handle = OpenThread(SYNCHRONIZE | THREAD_SUSPEND_RESUME |

THREAD_GET_CONTEXT,
+                            FALSE, thread->tid);
      } else {
          handle = NULL;
      }