Jacek Caban jacek@codeweavers.com writes:
On 03.08.2017 09:48, Alexandre Julliard wrote:
Jacek Caban jacek@codeweavers.com writes:
@@ -256,6 +256,7 @@ static void read_reply_data( void *buffer, size_t size ) if (!ret) break; if (errno == EINTR) continue; if (errno == EPIPE) break;
if (errno == EFAULT && virtual_uninterrupted_write_memory( buffer, NULL, size ) >= size) continue; server_protocol_perror("read");You can't enter a critical section in the middle of a server call.
Oh, right, we could need to do another server call there. Unless I'm missing something, it should work fine for platforms that have optimized CS implementation (Linux and Mac) as they don't do server calls. We could limit effect of the patch to those platform, but that would be ugly.
Could we just use pthread mutexes for server_enter_uninterrupted_section/server_leave_uninterrupted_section instead of critical sections?
I don't think you should be doing that sort of thing inside server calls at all. Some sort of lock/unlock_buffer methods that could be wrapped around server calls where necessary would probably work better.