| ★ wanayoo — archive 1999 http://www.kernelnotes.org/patch/21-p0199.html | Nouvelle recherche | Portail wanayoo |
|
Robert Krawitz (rlk@tiac.net) Thu Jan 16 11:48:00 1997 |
[Home] [Linux 2.0] [Linux 2.1] [Information] [Software] [Distributions] [Links] |
|
The testing procedure I used was as follows:
- Build the kernel with make -j. As of 1.3.90, this works.
- Run 2 and then 7 cvs diff commands concurrently, and check for nulls in the output. This test always failed with all previous versions of the patch. It passed cleanly with this latest patch.
This patch will do no good (and will probably hurt performance) on a 486 or lower. It will do no good on a crippled Pentium with a 32-bit memory bus (if your motherboard can accept a single 72-pin SIMM), but it may or may not hurt performance. It will do no good on a Nexgen or other processor with no FPU (it may or may not work, and will surely have terrible performance). It may or may not be effective on any other processor. You may compare the results of the Byte benchmarks with the patch with the results without the patch. As you'll note, the memcpy patch improves kernel performance by roughly 15%.
There has been a report that it hurt X performance severely on a system with an S3 video card, but there have been no other reports of this (there has since been a report from another S3 user who has not reported any problems). It is known that S3 (and some other) video cards can cause system crashes if /proc/mem is read with this routine (the various S3 chips apparently can't handle this, a fix is in the works). Finally, it is known that this does not work on an SMP. There is as yet no fix for the problem, although it's understood what's wrong.
An experimental version that handles this is here. This version is for 2.1.9 or higher (it was prepared against 2.1.10).