> -----Original Message----- > From: Stephen Hemminger [mailto:stephen@xxxxxxxxxxxxxxxxxx] > Sent: Friday, January 27, 2017 10:36 AM > To: KY Srinivasan <kys@xxxxxxxxxxxxx> > Cc: Haiyang Zhang <haiyangz@xxxxxxxxxxxxx>; > devel@xxxxxxxxxxxxxxxxxxxxxx; Stephen Hemminger > <sthemmin@xxxxxxxxxxxxx> > Subject: Re: [PATCH 07/14] vmbus: remove conditional locking of > vmbus_write > > On Fri, 27 Jan 2017 18:08:35 +0000 > KY Srinivasan <kys@xxxxxxxxxxxxx> wrote: > > > > -----Original Message----- > > > From: Stephen Hemminger [mailto:stephen@xxxxxxxxxxxxxxxxxx] > > > Sent: Monday, January 23, 2017 5:40 PM > > > To: KY Srinivasan <kys@xxxxxxxxxxxxx>; Haiyang Zhang > > > <haiyangz@xxxxxxxxxxxxx> > > > Cc: devel@xxxxxxxxxxxxxxxxxxxxxx; Stephen Hemminger > > > <sthemmin@xxxxxxxxxxxxx> > > > Subject: [PATCH 07/14] vmbus: remove conditional locking of > vmbus_write > > > > > > All current usage of vmbus write uses the acquire_lock flag, therefore > > > having it be optional is unnecessary. This also fixes a sparse warning > > > since sparse doesn't like when a function has conditional locking. > > > > In order to avoid cross-tree dependency, I got this functionality into Greg's > > tree first. I plan to submit patches to netvsc that will use this functionality. > > As you know, we hold a higher level lock in the protocol stack when we > send on > > a sub-channel. This will avoid an unnecessary spin lock round trip in the > data path. > > > > Regards, > > > > K. Y > > Conditional locking is in general a bad idea because it breaks static analysis > tools like > sparse. In order to have a non-locking code path then it is > better to have a locked and unlocked version of the same functions. Agreed. That said, I think this pattern is used in the kernel in many places. > > Also, in networking receive path is not specially single threaded. With my > per-channel > tasklet (and later NAPI), the each channel's receive path would be single > threaded > but there are still races possible in the shutdown logic (as written). Longer > term > it would be best if all vmbus events were handled per-channel without lock, > ie > networking is not a special case. The receive path is single threaded since the channel is bound to a CPU and all receive interrupts are delivered on the affinitized CPU. Indeed, on the read side we don't acquire any lock on the ring buffer (see hv_ringbuffer_read). The optimization I wanted to implement is on the write side. The sub-channel send code in netvsc is already serialized by a lock acquired at the upper level and this is what I wanted to take advantage of. > > Probably best to figure out how to make all VMBUS ring access lockless, ie it > is > callers responsibilty. Storage and networking are multi-queue already. This is probably the best way forward. Regards, K. Y _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
