On Wed, Mar 21, 2018 at 03:13:25PM -0500, Christopher Lameter wrote: > On Wed, 21 Mar 2018, Jason Gunthorpe wrote: > > > On Wed, Mar 21, 2018 at 01:04:45PM -0500, Christopher Lameter wrote: > > > On Wed, 21 Mar 2018, Jason Gunthorpe wrote: > > > > > > > I think we are going to add a lock, it should cover all readers > > > > too. There is no real reason not too that I can see. > > > > > > The general approach in the kernel is to not use locks for access to > > > statistics and recognize that they are snapshots and may be inaccurate > > > because unserialized things happened after or at the time of the snapshot. > > > > Maybe, but other places aren't recording their statistics in u64's > > then, there is no way to do that without locks portably. > > Umm... Ok so we are concerned about 32 bits here? True the core kernel > uses unsigned long for these counters. Sort of. For core kernel code like this I certainly don't want to see us cavalierly break things to solve imaginary problems. In this case I can't see a real performance problem here.. The extra two sysfs that need locking are not used frequently I think, and I doubt much software is polling sysfs files multi-threaded at a high rate of speed. That is a pretty nonsense thing to do. Jason -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
