> -----Original Message----- > From: KY Srinivasan > Sent: Tuesday, March 22, 2016 7:37 AM > To: 'Vitaly Kuznetsov' <[email protected]> > Cc: [email protected]; [email protected]; > [email protected]; [email protected]; [email protected]; > [email protected]; [email protected] > Subject: RE: [PATCH 3/5] Drivers: hv: vmbus: Fix a bug in > hv_need_to_signal_on_read() > > > > > -----Original Message----- > > From: Vitaly Kuznetsov [mailto:[email protected]] > > Sent: Tuesday, March 22, 2016 2:56 AM > > To: KY Srinivasan <[email protected]> > > Cc: [email protected]; [email protected]; > > [email protected]; [email protected]; [email protected]; > > [email protected]; [email protected] > > Subject: Re: [PATCH 3/5] Drivers: hv: vmbus: Fix a bug in > > hv_need_to_signal_on_read() > > > > KY Srinivasan <[email protected]> writes: > > > > >> -----Original Message----- > > >> From: Vitaly Kuznetsov [mailto:[email protected]] > > >> Sent: Monday, March 21, 2016 1:19 AM > > >> To: KY Srinivasan <[email protected]> > > >> Cc: [email protected]; [email protected]; > > >> [email protected]; [email protected]; [email protected]; > > >> [email protected]; [email protected] > > >> Subject: Re: [PATCH 3/5] Drivers: hv: vmbus: Fix a bug in > > >> hv_need_to_signal_on_read() > > >> > > >> "K. Y. Srinivasan" <[email protected]> writes: > > >> > > >> > We need to issue a full memory barrier prior making a signalling > > >> > decision. > > >> > > >> Any reason this should be mb()? This is pretty strong and will probably > > >> lead to performace regression ... and, btw, we have another mb() in > > >> hv_ringbuffer_read(). > > >> > > >> Could you please describe the scenarion you're trying to protect against > > >> so we could search for a better solution? > > > > > > If the reading of the pend_sz (in the function > > hv_need_to_signal_on_read) > > > were to be reordered and read before we commit the new read index > we > > could > > > have a problem. > > > > > > If these are two reads we can add a lightweight barrier just preventing > > compiler from reordering (e.g. smp_rmb()), right? > > > > > If the host were to set the pending_sz after we have sampled > pending_sz > > > and go to sleep before we commit the read index, we could miss sending > > > the interrupt. > > > > so write and then we read and we need to prevent reordering... not sure > > how to get rid on mb() then ... > > The other memory barrier in the function (prior to writing the read index) > has been there forever and I am not sure why that needs to be a full barrier. > I feel a read barrier should suffice.
I may also look at restructuring these APIs to not always check for signaling. I will experiment with that scheme to minimize the barrier calls. K. Y _______________________________________________ devel mailing list [email protected] http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
