Date: Thu, 15 Jan 2009 13:31:49 -0600 From: Dan Nelson <dnelson@allantgroup.com> To: Michael Powell <nightrecon@verizon.net> Cc: freebsd-questions@freebsd.org Subject: Re: kernel configuration Message-ID: <20090115193147.GA61100@dan.emsphone.com> In-Reply-To: <gknt72$39i$1@ger.gmane.org> References: <496E06D1.2070706@gmail.com> <20090114181522.GB4487@aurora.oekb.co.at> <B4DD918A-6026-4206-9C14-70DCD0028CFB@mac.com> <gknt72$39i$1@ger.gmane.org>
next in thread | previous in thread | raw e-mail | index | archive | help
In the last episode (Jan 15), Michael Powell said: > Chuck Swiger wrote: > > While it is true that you can comment out all but i686 and get a > > working kernel, you will experience reduced performance. There are > > a number of low-level assembly routines (cf sys/i386/i386/support.s > > such as i586_bcopy) that are conditionalized off of I586_CPU only, > > even though they provide an advantage on i686 platforms also. > > Thank you very kindly for this info tidbit. In my ignorance I had > routinely built my kernels with only I686_CPU. Since the few boxen I > have are all downlevel every niche I can squeak out some performance > is sought after. Thanks indeed! Actually, those functions are only enabled if the CPU is truly a 586-class processor. See /sys/i386/isa/npx.c , the npx_attach() function. There is a test for cpu_class==CPUCLASS_586, while most modern CPUs are CPUCLASS_686. -- Dan Nelson dnelson@allantgroup.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20090115193147.GA61100>