Date: Thu, 23 Jan 2014 14:52:41 -0500 From: John Baldwin <jhb@freebsd.org> To: Luigi Rizzo <rizzo@iet.unipi.it> Cc: freebsd-current@freebsd.org, FreeBSD Current <current@freebsd.org> Subject: Re: possible selrecord optimization ? Message-ID: <201401231452.41509.jhb@freebsd.org> In-Reply-To: <20140123003948.GC292@onelab2.iet.unipi.it> References: <CA%2BhQ2%2BhW4_8tkCqyUWUWR_VV%2B6Jp=t0XzVE5kaWFz=SKDd2bow@mail.gmail.com> <201401221429.56745.jhb@freebsd.org> <20140123003948.GC292@onelab2.iet.unipi.it>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wednesday, January 22, 2014 7:39:48 pm Luigi Rizzo wrote: > On Wed, Jan 22, 2014 at 02:29:56PM -0500, John Baldwin wrote: > > On Tuesday, January 21, 2014 9:25:27 pm Luigi Rizzo wrote: > > > Looking at how selrecord() / selwakeup() and their Linux counterparts > > > poll_wait() and wake_up() are used, i noticed the following: > .... > > > I wonder if we could use the same optimization as Linux: > > > as soon as pollscan/selscan detects a non-blocking fd, > > > make selrecord a no-op (which is probably as simple > > > as setting SELTD_RESCAN; and since it only goes up > > > we do not need to lock to check it). > > > > Yes, I think this would work fine. I think setting SELTD_RESCAN as a way to > > do it is fine as well. > > excellent, thanks. > > I also have two related questions: > > 1. why isn't the struct mtx part of the struct selinfo instead > of being grabbed from the mtxpool_select ? I think this is because there is no selinfo_init() and no selinfo_destroy(), so there is no way to manage the lifetime of the mutex were it embedded into the structure. Also, if there are a lot of these structures, but only a subset of them are ever accessed, then a smaller set of locks that are hashed onto the structures may work just fine without introducing extra contention (but with the benefit of saving space in the structures). > 2. am i correct that we do need to protect concurrent invocations > of selrecord() on the same selinfo because mtx_pool_find() > return the same mutex for a given struct selinfo ? If you mean 'do not need', yes. mtx_pool_find() does a hash on the address, so it will always return the same lock for a given selinfo, and no external locking should be needed by callers. However, callers often need to check other state for which they need to hold a lock anyway. OTOH, if, for example, socket buffer locks were reader/writer locks, sopoll_generic() could use read locks on the socket buffers even while calling selrecord(). > In case, any objections if i add some comments to the code > to explain the above ? Not from me. :) -- John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201401231452.41509.jhb>