Date: Wed, 14 Dec 2005 23:05:06 -0500 From: Jonathan Noack <noackjr@alumni.rice.edu> To: Scott Long <scottl@samsco.org> Cc: stable@freebsd.org Subject: Re: kernel cpu entries Message-ID: <43A0EB72.8060908@alumni.rice.edu> In-Reply-To: <43A0E916.7070204@samsco.org> References: <20051215002618.B4D3B5D07@ptavv.es.net> <43A0E607.2030101@alumni.rice.edu> <43A0E916.7070204@samsco.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Scott Long wrote: > Jonathan Noack wrote: >> Kevin Oberman wrote: >>> Scott Long wrote: >>>> Also, taking out CPU_I586 is usually a bad idea. It offers no >>>> performance penalties (unlike CPU_I386 and maybe CPU_I486), but >>>> enables things like optimized bcopy. >>> >>> >>> Ahh, This is the sort of thing I never realized. Is there anything in >>> the handbook that covers this? I had always been under the impression >>> that CPU_I686 enabled all things that the 686 was capable of. I will >>> build a new kernel to add that back in. >> >> >> From tuning(7): >> ************************************************** >> There are a number of *_CPU options that can be commented out. If you >> only want the kernel to run on a Pentium class CPU, you can easily >> remove I486_CPU, but only remove I586_CPU if you are sure your CPU is >> being recognized as a Pentium II or better. Some clones may be >> recognized as a Pentium or even a 486 and not be able to boot without >> those options. If it works, great! The operating system will be able >> to better use higher-end CPU features for MMU, task switching, >> timebase, and even device operations... >> ************************************************** >> >> From /sys/i386/conf/NOTES: >> ************************************************** >> # You must specify at least one CPU (the one you intend to run on); >> # deleting the specification for CPUs you don't need to use may make >> # parts of the system run faster. >> ************************************************** >> >> From npx(4) (also see /sys/i386/i386/support.s): >> ************************************************** >> The NPX registers are normally used to optimize copying and zeroing >> when all of the following conditions are satisfied: >> 1. cpu I586_CPU is an option >> ... >> Then copying and zeroing using the NPX registers is normally 30-100% >> faster. >> ************************************************** >> >> All is rosy until you see that I586_CPU looks like a loss for blowfish >> (if you have an i686 CPU): >> /sys/crypto/blowfish/arch/i386/bf_enc.S >> >> As I use AES, I guess I586_CPU is a win for me. Despite this, I think >> it makes the most sense for I686_CPU to enable the optimized bcopy if >> it really is a win for i686 CPUs. > > I agree, but frankly I've been loath to touch it out of pure fear of the > correctness geeks. I know that if I go near it, someone will point out > that it's not 100% correct in all cases of some buggy i686 derivative > that hasn't been sold since 1998, and therefore it's better to just > leave it alone and satify that .00001% of the problem. Or, the > alternate scenario is that people will moan that we should be using > SSE instead, and that any change that doesn't involve SSE is wrong and > a waste of time. Then a meta-argument will break out over SSE vs SSE2 > vs 3DNow, and how again some buggy derivative chip can't use it and > can't be detected or worked around. I make my peace by just remembering > to leave CPU_I586 enabled on all of my local systems =-) It's even a project idea: http://www.freebsd.org/projects/ideas/#p-memcpy -Jonathan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?43A0EB72.8060908>