Messages in this thread |  | | Subject | Re: Crap: CONFIG_RANDOM and TCP FILTER | Date | Mon, 20 May 1996 15:11:50 -0600 | From | Warner Losh <> |
| |
: If you dont have 16Mb saving 16K is _VERY_ useful, thats 4 pages of user : space. Its also important people do keep trying to strip down the kernel to : keep its size good. We need more cleaner smaller drivers like Paul Gortmakers : dumber IDE/MFM/RLL driver.
The floppy driver could stand a good fresh rewrite. It is too big and twisted right now. Heck, OpenBSD's floppy driver is 1/2 the size of Linux's in terms of lines of code and 1/8 in terms of size(1) of the .o. If people really want to do something about kernel bloat, this might be fruitful as it will help on just about every machine the Linux is running on, even the Sparc and MIPS ports! The last MIPS compile I did it was over 56k in size! Don't you think that might be a little on the large size? The OpenBSD/Pica (mips) fd.o is only 7k in size using the same compiler rev for both!!! The FreeBSD/intel binary size is 10k, so I think that at least 40k if not 50k could be saved if a good, sleek floppy driver fully integrated into the buffer cache was produced. I don't have a recent intel kernel build handy, or I'd be more specific.
All of this noise about a mere 16k when there are other, larger poles in the tent is downright silly. Make sure that you get the right things removed from the kernel first then come back later if the 16k pole is still too big. Close analysis of the kernel will likely reveal other areas where the code is too big and a win could happen by simplification.
Warner
P.S. I'm not picky on the floppy driver, per se, in this message. Just the concept of looking at some random part of the kernel and saying that it is too big when there are other, larger parts of the kernel that are even bigger and could easily be slimmed down. The floppy driver in FreeBSD/OpenBSD needs to be rewritten as well and I'm not sure if the same level of functionality is present in both or not. A factor of 8 is kinda hard to ignore :-)
P.P.S. Please don't take this as a Linux flame. All OSes that I play with have this problem in one area or another. It seems that very few people these days are taking good, hard looks at code size on a regular and recurring basis. Look at Solaris or VMS or FreeBSD or Windows or whatever and you'll see similar things happening.
|  |