Messages in this thread | | | Date | Tue, 23 Dec 1997 22:58:39 -0600 (CST) | From | Phil Brutsche <> | Subject | Re: [PATCH] Intel clock speed detection |
| |
On another note, I just looked at your patch. It uses the rdtsc instruction; this instruction does not exist on anything less than a Pentium. I'm not sure what the kernel would do on '486 or '386, but it would'nt be pretty. A better solution for this would be to time the bsf instruction on those chips. I have source in C and assembly if you need help.
On Tue, 23 Dec 1997, Nicholas J. Leon wrote:
> Fellow kernel enthusiasts - > > I've been playing with routines to determine actual cpu speed on Intel > chips; hoping to replace the "75+" notation with something accurate. > > Using some assembly routines from a patch I acquired a few months ago, > whose author I have completely forgotten, I've made some preliminary > patches against 2.1.75. Here is what I did: > > . added a field called "x86_mhz" to cpuinfo_x86 > . added a function called "get_cpu_speed" to setup.c > . calling get_cpu_speed from identify_cpu() in setup.c > . added reporting of the speed to print_cpu_info() and get_cpuinfo() > in setup.c > > Here are the problems I know I haven't addressed: > > . does my mhz detection routine work reliably (ie, at all) > with all non-intel cpu's? > . what about SMP machines? > . slight variation of reported mhz (132 vs 133) > . struct cpu_model_info now has redundant elements of "Pentium" > (since I removed the 60/66, 75+). I'm not sure how these are > referenced in the first place, so I didn't compact them together > (as they should be). > > The first I am hoping that some people out there will try my patch on > their AMD and Cyrix systems and let me know. The second, uhhhhh, well, > I'm over my head with this one. I would certainly appreciate any and > all help that might come my direction. Regardless of knowledge, > though, I have no practical way of testing SMP's reaction to my > patch. So I must rely on others for this. > > I know that in the past, Linus has refused to include clock speed > detecting into the kernel. From what I understand, the original patch > did the calculations on the fly when you examined /proc/cpuinfo and > this was the reason for the refusal. My patch doesn't. It calculates > the rate upon boot and stores the information into > cpuinfo_x86. Hopefully, this will allow its inclusion into the > mainstream kernel once all the issues have been taken care of. > > Please, take the time and look at what I've done. I know its not > anything amazing, but since the kernel isn't written in Perl, there's > not a lot I can contribute :) This, however, might be one of those > things I _can_ do. > > Thank you for your time! > > [ patch attached ] > > > ___________________________________________________________________________ > > simple is elegant nicholas@binary9.net > ___________________________________________________________________________ >
---------------------------------------------------------------------- Phil Brutsche
"Be of stout heart, Number One. We've handled the Borg. We can certainly handle Admiral Jellico." - Jean-Luc Picard
----------------------------------------------------------------------
| |