[07/21] add assertions on the owner of a QemuMutex

These are already present in the Win32 implementation, add them to
the pthread wrappers as well.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 qemu-thread-posix.c |   27 +++++++++++++++++++++++++--
 qemu-thread-posix.h |    1 +
 2 files changed, 26 insertions(+), 2 deletions(-)

On 2011-02-21 09:43, Paolo Bonzini wrote:
> These are already present in the Win32 implementation, add them to
> the pthread wrappers as well.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>  qemu-thread-posix.c |   27 +++++++++++++++++++++++++--
>  qemu-thread-posix.h |    1 +
>  2 files changed, 26 insertions(+), 2 deletions(-)
> 
> diff --git a/qemu-thread-posix.c b/qemu-thread-posix.c
> index 28b3f80..2176f81 100644
> --- a/qemu-thread-posix.c
> +++ b/qemu-thread-posix.c
> @@ -16,9 +16,12 @@
>  #include <time.h>
>  #include <signal.h>
>  #include <stdint.h>
> +#include <assert.h>
>  #include <string.h>
>  #include "qemu-thread.h"
>  
> +static pthread_t pthread_null;
> +
>  static void error_exit(int err, const char *msg)
>  {
>      fprintf(stderr, "qemu: %s: %s\n", msg, strerror(err));
> @@ -28,8 +31,13 @@ static void error_exit(int err, const char *msg)
>  void qemu_mutex_init(QemuMutex *mutex)
>  {
>      int err;
> +    pthread_mutexattr_t mutexattr;
>  
> -    err = pthread_mutex_init(&mutex->lock, NULL);
> +    mutex->owner = pthread_null;
> +    pthread_mutexattr_init(&mutexattr);
> +    pthread_mutexattr_settype(&mutexattr, PTHREAD_MUTEX_ERRORCHECK);
> +    err = pthread_mutex_init(&mutex->lock, &mutexattr);
> +    pthread_mutexattr_destroy(&mutexattr);
>      if (err)
>          error_exit(err, __func__);
>  }
> @@ -48,13 +56,20 @@ void qemu_mutex_lock(QemuMutex *mutex)
>      int err;
>  
>      err = pthread_mutex_lock(&mutex->lock);
> +    mutex->owner = pthread_self();
>      if (err)
>          error_exit(err, __func__);
>  }
>  
>  int qemu_mutex_trylock(QemuMutex *mutex)
>  {
> -    return pthread_mutex_trylock(&mutex->lock);
> +    int err;
> +    err = pthread_mutex_trylock(&mutex->lock);
> +    if (err == 0) {
> +        mutex->owner = pthread_self();
> +    }
> +
> +    return !!err;
>  }
>  
>  static void timespec_add_ms(struct timespec *ts, uint64_t msecs)
> @@ -85,6 +100,11 @@ void qemu_mutex_unlock(QemuMutex *mutex)
>  {
>      int err;
>  
> +    /* An EDEADLOCK would arrive after we reset the owner.  So this
> +       assert is for ease of debugging (it lets you see what is the
> +       actual owner.  */

Don't get this. Why do you want to avoid the proper error detection of
pthread?

> +    assert(pthread_equal(mutex->owner, pthread_self()));
> +    mutex->owner = pthread_null;
>      err = pthread_mutex_unlock(&mutex->lock);
>      if (err)
>          error_exit(err, __func__);
> @@ -130,7 +150,10 @@ void qemu_cond_wait(QemuCond *cond, QemuMutex *mutex)
>  {
>      int err;
>  
> +    assert(pthread_equal(mutex->owner, pthread_self()));
> +    mutex->owner = pthread_null;
>      err = pthread_cond_wait(&cond->cond, &mutex->lock);
> +    mutex->owner = pthread_self();
>      if (err)
>          error_exit(err, __func__);
>  }
> diff --git a/qemu-thread-posix.h b/qemu-thread-posix.h
> index 7af371c..11978db 100644
> --- a/qemu-thread-posix.h
> +++ b/qemu-thread-posix.h
> @@ -4,6 +4,7 @@
>  
>  struct QemuMutex {
>      pthread_mutex_t lock;
> +    pthread_t owner;
>  };
>  
>  struct QemuCond {

You said that you want to add owner tracking for assertion in
cond_signal/broadcast. That's OK. But the assertions in the mutex layer
are redundant for PTHREAD_MUTEX_ERRORCHECK - unless I'm missing
something now.

Jan

On 02/21/2011 10:50 AM, Jan Kiszka wrote:
>> >  +    /* An EDEADLOCK would arrive after we reset the owner.  So this
>> >  +       assert is for ease of debugging (it lets you see what is the
>> >  +       actual owner.  */
>
> Don't get this. Why do you want to avoid the proper error detection of
> pthread?

Because by the time you get to error_exit mutex->owner has been NULL-ed 
out already.  So it doesn't help to put a breakpoint on error_exit, you 
cannot find out which thread was the owner.

> But the assertions in the mutex layer
> are redundant for PTHREAD_MUTEX_ERRORCHECK - unless I'm missing
> something now.

Yes, but tracking the mutex's owner gives a bit more specific 
information when an error happens even for mutexes.  I removed them from 
lock/trylock, but for unlock it's already too late when the error happens.

Paolo

On 2011-02-21 11:15, Paolo Bonzini wrote:
> On 02/21/2011 10:50 AM, Jan Kiszka wrote:
>>>>  +    /* An EDEADLOCK would arrive after we reset the owner.  So this
>>>>  +       assert is for ease of debugging (it lets you see what is the
>>>>  +       actual owner.  */
>>
>> Don't get this. Why do you want to avoid the proper error detection of
>> pthread?
> 
> Because by the time you get to error_exit mutex->owner has been NULL-ed 
> out already.  So it doesn't help to put a breakpoint on error_exit, you 
> cannot find out which thread was the owner.

That's easy, "p <my_mutex>" will tell (the structure contains the
owner's tid).

And for debugging invalid mutex_unlock calls, it's more interesting to
track the call path of that thread which incorrectly claimed to hold the
lock.

Jan

On 02/21/2011 11:22 AM, Jan Kiszka wrote:
> That's easy, "p<my_mutex>" will tell (the structure contains the
> owner's tid).
>
> And for debugging invalid mutex_unlock calls, it's more interesting to
> track the call path of that thread which incorrectly claimed to hold the
> lock.

Ok, will remove.

Paolo