> On Mon, Mar 27, 2017 at 05:39:27PM +0200, Mateusz Guzik wrote: > > The previous patch replaced multiple reads of the global var with just > > one read and had the result stored in a local variable, which then is > > read multiple times. Even though the compiler ended up emitting one read > > of the global, it perfectly legal to emit several, thus the bug can still > > reappear. This can be prevented as described above. > > It is not a problem when the compiler creates multiple read > instructions for syslogf as long no function call is between them. > Nothing else can change the content of syslogf in parallel, we are > running with the big kernel lock. > > It would be a problem if the compiler would do a read from the > global syslogf variable before and after a sleep. But that would > be illegal in C. > > When we will unlock this code for multi processor, we need lock, > memory barrier, per cpu memory, or whatever like you suggested. > But that does not happen now.
This really has nothing to do with the biglock. As the biglock gets pushed out of this layer, other locks will be added to ensure that the VFS functions can be safely call, those will satisfy the conditions. There's no point creating needless contortions.
