Re: [2.6 patch] schedule obsolete OSS drivers for removal

[Nix <nix@esperi.org.uk>]

On 30 Oct 2005, Adrian Bunk said:
> E.g. it's clear that unused code or unused EXPORT_SYMBOL's in the kernel 
> are bloat, so I am working on eliminating such bloat.

And thank you very much for that!

What's most notable to me is a substantial cross-arch reduction in .data
space consumed. This non-FUSE-2.6.13 versus FUSE-2.6.14 UltraSPARC
comparison shows this effect clearly:

-rwxr-xr-x  1 root root 3460304 Oct 13 00:55 vmlinux-2.6.13.4
-rwxr-xr-x  1 root root 3190448 Oct 29 17:18 vmlinux-2.6.14

                         2.6.13.4     2.6.14 
section                      size       size    diff
.text                     2330144    2341280  +10236
.rodata                    198361     199009    +648
.pci_fixup                   1536       1536       0
__ksymtab                   32352      32336     -16
__ksymtab_gpl                3536       4048    +512
__kcrctab                       0          0       0
__kcrctab_gpl                   0          0       0
__ksymtab_strings           44688      45424    +736
__param                      1640       1640       0
.data                      669288     387576 -281712
.data.cacheline_aligned       704        704       0
.data.read_mostly             428      13144  -12716
.fixup                      18256      18268     +12
__ex_table                  15760      15768      +8
.init.text                 109728      94528  -15200
.init.data                   7280       8712   +1432
.init.setup                   792        816     +24
.initcall.init                784        808     +24
.con_initcall.init             16         16       0
.security_initcall.init         0          0       0
.init.ramfs                   134        134       0
.bss                       226528     226152    -376
.note.GNU-stack                 0          0       0
__ex_table.1                  216        216       0
__ex_table.2                   16        496    +480
Total                     3662187    3392611 -269576

On a similarly-configured Athlon the difference in .data is a little
less pronounced, but still there.

UltraSPARC 2.6.13.4 versus 2.6.14 at boot:

Memory: 511576k available (2280k kernel code, 936k data, 128k init) [fffff80000000000,0000000037ebc000]
Memory: 511848k available (2288k kernel code, 672k data, 112k init) [fffff80000000000,0000000037ebc000]

(that's 264K, the same as the .data size difference above)

Athlon 2.6.13.4 versus 2.6.14 at book:

Memory: 774828k/786432k available (2794k kernel code, 11156k reserved, 1022k data, 148k init, 0k highmem)
Memory: 774996k/786432k available (2833k kernel code, 10988k reserved, 808k data, 152k init, 0k highmem)

(that's 214K different in .data although kernel code has grown a little
here, probably thanks to FUSE).

Most OSes cannot claim to shrink across upgrades. Linux, at least the
kernel, can. :)

-- 
`"Gun-wielding recluse gunned down by local police" isn't the epitaph
 I want. I am hoping for "Witnesses reported the sound up to two hundred
 kilometers away" or "Last body part finally located".' --- James Nicoll