From: Masahiro A. <m-...@aa...> - 2003-03-28 02:32:31
|
On Wed, 29 Jan 2003 11:35:32 +0900 Masahiro Abe <m-...@aa...> wrote: > I still haven't being able to make TLB-Miss handling speed-up part work on > my SolutionEngine. > I must do something else right now so my investigation pends now. I > really appreciate if anyone can help on this. If anyone can give more > info/suggestion/patch, I'll try that. Since then, Stuart let me see his entire source tree, and some missing parts are found. Now it seems to be working with those changes. Also, SUGIOKA-san generously gave me his version of modification. I also tried that and it seems to be working fine. I'm not sure which is better, so I've assembled two version of patches for TLB Miss speede-up and put in: <ftp://ftp.aandd.co.jp/pub/linuxsh/work/20030308/> There are 5 patches there. 1. diff-2.4.18-cvs.diff.bz2 is to apply changes to stock 2.4.18 kernel, and make it same as the "linux-2_4_18" tagged cvs source. 2. diff-2.4.18cvs-se.diff.bz2 is to apply to the "linux-2_4_18" tagged cvs source, and make it boot from CompactFlash on SolutionEngine 7750S. This is not needed for other platforms. 3. diff-2.4.18cvs-memcpy-20030308-1.diff.bz2 is to apply to the "linux-2_4_18" tagged cvs source, and includes Stuart's "memcpy speed-up" (with fixes that discussed later on the list) For "TLB Miss Handling optimization", there are two types: 4. diff-2.4.18cvs-tlb-sm-20030308-1.diff.bz2 includes Mr. Stuart Menefy's original patch plus some missing parts. 5. diff-2.4.18cvs-tlb-ts-20030308-1.diff.bz2 includes Mr. Toshinobu Sugioka's modified patch. I hope these files be tried by many of you and grows to good shape. It will be good if 4 and 5 be merged to be a definitive implementation for TLB Miss handling speed-up. (I can't do that by myself, because I don't understand the exact detail of those patches. At least for now.) --- By the way, I've also assembled patches to make RTLinux-3.2-pre1 work with above kernels. Those are put in: <ftp://ftp.aandd.co.jp/pub/linuxsh/rtlinux/work/20030327/> These are not "Real-TIme" yet, though. Timer latency still gets to ~500usec with heavy disk access. ================================= Masahiro ABE, A&D Co., Ltd. Japan |