From mboxrd@z Thu Jan 1 00:00:00 1970 Content-Type: multipart/mixed; boundary="===============9141689981562504977==" MIME-Version: 1.0 From: Walker, Benjamin Subject: Re: [SPDK] SPDK Event notifications Date: Thu, 11 Oct 2018 20:57:28 +0000 Message-ID: <7dd38725631db7b4856b6c4be3f1f69efa3927d8.camel@intel.com> In-Reply-To: CANvN+emwAZKsYdoWAo5zMS_T+1QOj71x2pxBhHfYOC_J1iuEGw@mail.gmail.com List-ID: To: spdk@lists.01.org --===============9141689981562504977== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Thu, 2018-10-11 at 22:13 +0200, Andrey Kuzmin wrote: > Just curious, how does SPDK NVMe driver handle AENs then (if at all), as > the problem statement above is pretty similar to what AEN aimed to addres= s? The driver just sends down an AEN command at start up which sits there unti= l an event happens. Users may register for notifications upon completion of that= AEN command. Anyone registered at the time that the AEN fires gets notified. Wh= en the driver sees the AEN completion, it automatically clears the event by se= nding the appropriate get log page command. That means that events will be thrown= away if no one is subscribed at the time they occur. AEN is similarly flawed in that someone must clear the event, which implies= that all clients are coordinating with one another somehow. If one client clears= the event, future clients won't be notified. In general this isn't a problem for NVMe because there is only one client (the device driver) and when that isn= 't connected the device is not powered. > Regards, > Andrey --===============9141689981562504977==--