Re: [PATCH v4 2/2] rcu: add uninit destructor for rcu

[Paolo Bonzini]

On 09/09/20 10:41, Stefan Hajnoczi wrote:
> On Tue, Sep 08, 2020 at 11:10:52PM +0800, Yonggang Luo wrote:
>> This is necessary if the pending  rcu calls are closing and removing
>> temp files. This also provide a function
>> void rcu_wait_finished(void);
>> to fixes test-logging.c test failure on msys2/mingw.
>> On windows if the file doesn't closed, you can not remove it.
>>
>> Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
>> ---
>>  include/qemu/rcu.h   |  5 +++++
>>  tests/test-logging.c |  2 ++
>>  util/rcu.c           | 37 ++++++++++++++++++++++++++++++++++++-
>>  3 files changed, 43 insertions(+), 1 deletion(-)
> Can the new drain_call_rcu() function be used? Maxim recently posted the
> following patch:
> https://patchew.org/QEMU/20200831150124.206267-1-mlevitsk@redhat.com/20200831150124.206267-3-mlevitsk@redhat.com/
> 
> Whether drain_call_rcu() or rcu_wait_finished() is used, please include
> a comment in the code that documents why the wait is necessary. For
> example, "qemu_log_close() uses RCU for its FILE pointer but Windows
> cannot remove open files, so we need to wait for RCU here".
> 
> Another option is to wait for RCU inside qemu_log_close() so that
> callers don't need to worry about this implementation detail:
> 
>   #ifdef _WIN32
>   /* Windows cannot remove open files so we need to wait for RCU here */
>   drain_call_rcu();
>   #endif
> 

In this case even synchronize_rcu() should be okay.

Paolo

signature.asc
Description: OpenPGP digital signature