* [ANNOUNCE] 2.6.0-tiny1 tree for small systems
@ 2003-12-27 21:56 Matt Mackall
2003-12-28 10:35 ` Willy Tarreau
2003-12-28 23:54 ` Mike Fedyk
0 siblings, 2 replies; 7+ messages in thread
From: Matt Mackall @ 2003-12-27 21:56 UTC (permalink / raw)
To: linux-kernel
This is the second release of the -tiny kernel tree. The aim of this
tree is to collect patches that reduce kernel disk and memory
footprint as well as tools for working on small systems. Target users
are things like embedded systems, small or legacy desktop folks, and
handhelds.
Latest release includes:
- sync against 2.6.0
- fix for CONFIG_SYSCTL compilation
- Jeff Garzik's netdrvr patchkit, along with latest netpoll/netconsole code
- kgdb and kgdb-over-ethernet debugging support
- "make checkstack" to find largest stack users
- script to estimate actual code size of inline functions (experimental)
- some other minor new tweaks
Thanks to Holger Shuring, Bill Irwin, and Zwane Mwaikamba for sending
me various tweaks.
The patch, currently against 2.6.0, can be found at:
http://selenic.com/tiny/2.6.0-tiny1.patch.bz2
http://selenic.com/tiny/2.6.0-tiny1-broken-out.tar.bz2
And while I'm at it, I'll include some demo logs:
[my tiny test config: ide + vt + ext2 + net ]
$ size vmlinux
text data bss dec hex filename
861316 76979 20524 958819 ea163 vmlinux
LILO boot: tiny mem=4m
Loading tiny.........
Uncompressing Linux... Ok, booting the kernel.
# cat /proc/meminfo
MemTotal: 2772 kB
MemFree: 1096 kB
Buffers: 64 kB
Cached: 896 kB
SwapCached: 0 kB
Active: 788 kB
Inactive: 392 kB
HighTotal: 0 kB
HighFree: 0 kB
LowTotal: 2772 kB
LowFree: 1096 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
Mapped: 616 kB
Slab: 356 kB
Committed_AS: 280 kB
PageTables: 48 kB
VmallocTotal: 1032168 kB
VmallocUsed: 0 kB
VmallocChunk: 1032168 kB
#
[rebooting with kmalloc auditing and kallsyms]
# cat /proc/kmalloc
total bytes allocated: 266848
slack bytes allocated: 37774
net bytes allocated: 145568
number of allocs: 732
number of frees: 282
number of callers: 71
lost callers: 0
lost allocs: 0
unknown frees: 0
total slack net alloc/free caller
256 203 256 8/0 alloc_vfsmnt+0x73
8192 3648 4096 2/1 atkbd_connect+0x1b
192 48 64 3/2 seq_open+0x10
12288 0 4096 3/2 seq_read+0x53
8192 0 0 2/2 alloc_skb+0x3b
960 0 0 10/10 load_elf_interp+0xa1
1920 288 0 10/10 load_elf_binary+0x100
320 130 0 10/10 load_elf_binary+0x1d8
192 48 96 6/3 request_irq+0x22
7200 1254 7200 75/0 proc_create+0x74
64 43 64 2/0 proc_symlink+0x40
4096 984 0 1/1 check_partition+0x1b
69632 0 45056 17/6 dup_task_struct+0x38
128 48 128 2/0 netlink_create+0x84
128 20 128 1/0 ext2_fill_super+0x2f
32 28 32 1/0 ext2_fill_super+0x385
32 31 32 1/0 ext2_fill_super+0x3b6
608 76 384 19/7 __request_region+0x18
64 32 64 2/0 rand_initialize_disk+0xd
8192 2016 8192 2/0 alloc_tty_struct+0x10
128 56 128 2/0 init_dev+0xba
128 56 128 2/0 init_dev+0xf3
128 0 128 2/0 create_workqueue+0x28
8960 1680 8960 70/0 tty_add_class_device+0x20
2048 960 2048 4/0 alloc_tty_driver+0x10
9280 2332 9280 4/0 tty_register_driver+0x2d
288 0 288 9/0 mempool_create+0x16
1280 196 1280 9/0 mempool_create+0x41
1536 384 1536 8/0 mempool_create+0x8f
64 28 64 1/0 kbd_connect+0x3e
928 348 0 29/29 kmem_cache_create+0x235
28288 1448 28288 81/0 do_tune_cpucache+0x2c
...
[counting inline function calls]
$ make 2>../inline-info
CHK include/linux/version.h
SPLIT include/linux/autoconf.h -> include/config/*
make[1]: `arch/i386/kernel/asm-offsets.s' is up to date.
CHK include/asm-i386/asm_offsets.h
CC init/main.o
...
LD arch/i386/boot/compressed/vmlinux
OBJCOPY arch/i386/boot/vmlinux.bin
BUILD arch/i386/boot/bzImage
Root device is (3, 3)
Boot sector 512 bytes.
Setup is 2538 bytes.
System is 485 kB
Kernel: arch/i386/boot/bzImage is ready
$ scripts/count-inlines ../inline-info
2682 current_thread_info (485 in *.c)
calls:
callers: get_current(1764) <other>(276) copy_from_user(135)
copy_to_user(131) test_thread_flag(36) tcp_put_port(23)
tcp_set_state(22) verify_area(21) TLan_SetTimer(10)
tcp_enter_memory_pressure(6) set_thread_flag(6) sock_orphan(5)
inet_putpeer(5) dev_kfree_skb_any(5) clear_thread_flag(5)
inet_getid(3) sock_graft(2) igmp_mcf_get_first(2) icmp_xmit_lock(2)
csum_and_copy_to_user(2) test_and_clear_thread_flag(1)
tcp_v4_lookup(1) neigh_update_hhs(1) ip_finish_output2(1) gfp_any(1)
fn_flush_list(1) do_kdsk_ioctl(1) do_kdgkb_ioctl(1)
do_kbkeycode_ioctl(1) do_getname(1) alloc_buf(1) activate_task(1)
1764 get_current (1494 in *.c)
calls: current_thread_info(2682)
callers: <other>(417) capable(159) flush_tlb_page(11) flush_tlb_mm(10)
do_trap(10) on_sig_stack(9) sas_ss_flags(8) find_process_by_pid(6)
check_sticky(5) flush_tlb_range(4) walk_init_root(3) scm_send(3)
current_is_pdflush(3) current_is_kswapd(3) rwsem_down_failed_common(2)
kmem_freepages(2) get_sigframe(2) follow_dotdot(2) do_mmap2(2)
do_follow_link(2) __vfs_follow_link(2) __exit_mm(2) scm_check_creds(1)
save_i387_fsave(1) restore_i387_fsave(1) read_zero_pagealigned(1)
handle_group_stop(1) good_sigevent(1) get_close_on_exec(1)
fork_traceflag(1) ext2_init_acl(1) exec_permission_lite(1) dup_mmap(1)
do_tty_write(1) de_thread(1) copy_signal(1) copy_sighand(1) copy_fs(1)
cap_set_all(1) cap_netlink_send(1) cap_emulate_setxuid(1)
arch_get_unmapped_area(1)
...
[top stack users]
$ make checkstack
objdump -d vmlinux | perl scripts/checkstack.pl i386
1492 ide_unregister
1428 huft_build
1320 inflate_dynamic
1168 inflate_fixed
552 ip_setsockopt
460 tcp_v4_conn_request
456 tcp_check_req
448 tcp_timewait_state_process
440 ip_getsockopt
368 extract_entropy
320 do_execve
288 icmp_send
...
--
Matt Mackall : http://www.selenic.com : Linux development and consulting
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [ANNOUNCE] 2.6.0-tiny1 tree for small systems
2003-12-27 21:56 [ANNOUNCE] 2.6.0-tiny1 tree for small systems Matt Mackall
@ 2003-12-28 10:35 ` Willy Tarreau
2004-01-01 1:46 ` Matt Mackall
2003-12-28 23:54 ` Mike Fedyk
1 sibling, 1 reply; 7+ messages in thread
From: Willy Tarreau @ 2003-12-28 10:35 UTC (permalink / raw)
To: Matt Mackall; +Cc: linux-kernel
On Sat, Dec 27, 2003 at 03:56:06PM -0600, Matt Mackall wrote:
> This is the second release of the -tiny kernel tree. The aim of this
> tree is to collect patches that reduce kernel disk and memory
> footprint as well as tools for working on small systems. Target users
> are things like embedded systems, small or legacy desktop folks, and
> handhelds.
Hi Matt,
This looks very interesting. I could produce a small "bloated" kernel
which I use for kexec and remote recovery on a VIA C3-based system :
text data bss dec hex filename
1328734 118955 52016 1499705 16e239 vmlinux
This was with gcc-3.3.1 which is more efficient than its predecessors
on -Os.
I changed -march=c3 to -march=i386, which generally gives me better
numbers :
text data bss dec hex filename
1287578 118955 52016 1458549 164175 vmlinux
=> this is 40 kB saved. I simply changed this :
--- ./arch/i386/Makefile Sun Dec 28 10:35:23 2003
+++ ./arch/i386/Makefile Sun Dec 28 11:16:10 2003
@@ -44,7 +44,7 @@
cflags-$(CONFIG_MWINCHIPC6) += $(call check_gcc,-march=winchip-c6,-march=i586)
cflags-$(CONFIG_MWINCHIP2) += $(call check_gcc,-march=winchip2,-march=i586)
cflags-$(CONFIG_MWINCHIP3D) += $(call check_gcc,-march=winchip2,-march=i586)
-cflags-$(CONFIG_MCYRIXIII) += $(call check_gcc,-march=c3,-march=i486) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
+cflags-$(CONFIG_MCYRIXIII) += -march=i386 $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
cflags-$(CONFIG_MVIAC3_2) += $(call check_gcc,-march=c3-2,-march=i686)
CFLAGS += $(cflags-y)
So it might be interesting to have an option to drop back to the smallest
common arch during compilation. Note that this is not the same as choosing
i386 as the target system, since this kernel still has C3 features. Perhaps
the simplest and most portable way would be to create a new config entry
which would override default cflags ?
Cheers,
Willy
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [ANNOUNCE] 2.6.0-tiny1 tree for small systems
2003-12-27 21:56 [ANNOUNCE] 2.6.0-tiny1 tree for small systems Matt Mackall
2003-12-28 10:35 ` Willy Tarreau
@ 2003-12-28 23:54 ` Mike Fedyk
2003-12-29 0:31 ` William Lee Irwin III
1 sibling, 1 reply; 7+ messages in thread
From: Mike Fedyk @ 2003-12-28 23:54 UTC (permalink / raw)
To: Matt Mackall; +Cc: linux-kernel, wli
On Sat, Dec 27, 2003 at 03:56:06PM -0600, Matt Mackall wrote:
> This is the second release of the -tiny kernel tree. The aim of this
> tree is to collect patches that reduce kernel disk and memory
> footprint as well as tools for working on small systems. Target users
> are things like embedded systems, small or legacy desktop folks, and
> handhelds.
>
> Latest release includes:
> - "make checkstack" to find largest stack users
Maybe wli will be interested in this one since he has some stack shrinking
patches in his tree...
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [ANNOUNCE] 2.6.0-tiny1 tree for small systems
2003-12-28 23:54 ` Mike Fedyk
@ 2003-12-29 0:31 ` William Lee Irwin III
0 siblings, 0 replies; 7+ messages in thread
From: William Lee Irwin III @ 2003-12-29 0:31 UTC (permalink / raw)
To: mfedyk, Matt Mackall, linux-kernel
On Sat, Dec 27, 2003 at 03:56:06PM -0600, Matt Mackall wrote:
>> This is the second release of the -tiny kernel tree. The aim of this
>> tree is to collect patches that reduce kernel disk and memory
>> footprint as well as tools for working on small systems. Target users
>> are things like embedded systems, small or legacy desktop folks, and
>> handhelds.
>> Latest release includes:
>> - "make checkstack" to find largest stack users
On Sun, Dec 28, 2003 at 03:54:17PM -0800, Mike Fedyk wrote:
> Maybe wli will be interested in this one since he has some stack shrinking
> patches in his tree...
I'm already following this in general. I contributed a number of fixes
I've done for the 4K stack code over time at the time mpm originally
put -tiny together, though I think he's rearranged various things (e.g.
re-split the thing into 3 pieces) I can't be arsed to deal with.
-- wli
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [ANNOUNCE] 2.6.0-tiny1 tree for small systems
2003-12-28 10:35 ` Willy Tarreau
@ 2004-01-01 1:46 ` Matt Mackall
2004-01-03 16:50 ` Willy TARREAU
0 siblings, 1 reply; 7+ messages in thread
From: Matt Mackall @ 2004-01-01 1:46 UTC (permalink / raw)
To: Willy Tarreau; +Cc: linux-kernel
On Sun, Dec 28, 2003 at 11:35:00AM +0100, Willy Tarreau wrote:
> On Sat, Dec 27, 2003 at 03:56:06PM -0600, Matt Mackall wrote:
> > This is the second release of the -tiny kernel tree. The aim of this
> > tree is to collect patches that reduce kernel disk and memory
> > footprint as well as tools for working on small systems. Target users
> > are things like embedded systems, small or legacy desktop folks, and
> > handhelds.
>
> Hi Matt,
>
> This looks very interesting. I could produce a small "bloated" kernel
> which I use for kexec and remote recovery on a VIA C3-based system :
>
> text data bss dec hex filename
> 1328734 118955 52016 1499705 16e239 vmlinux
>
> This was with gcc-3.3.1 which is more efficient than its predecessors
> on -Os.
>
> I changed -march=c3 to -march=i386, which generally gives me better
> numbers :
>
> text data bss dec hex filename
> 1287578 118955 52016 1458549 164175 vmlinux
>
> => this is 40 kB saved. I simply changed this :
>
> +++ ./arch/i386/Makefile Sun Dec 28 11:16:10 2003
> @@ -44,7 +44,7 @@
> cflags-$(CONFIG_MWINCHIPC6) += $(call check_gcc,-march=winchip-c6,-march=i586)
> cflags-$(CONFIG_MWINCHIP2) += $(call check_gcc,-march=winchip2,-march=i586)
> cflags-$(CONFIG_MWINCHIP3D) += $(call check_gcc,-march=winchip2,-march=i586)
> -cflags-$(CONFIG_MCYRIXIII) += $(call check_gcc,-march=c3,-march=i486) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
> +cflags-$(CONFIG_MCYRIXIII) += -march=i386 $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
> cflags-$(CONFIG_MVIAC3_2) += $(call check_gcc,-march=c3-2,-march=i686)
>
> CFLAGS += $(cflags-y)
>
>
> So it might be interesting to have an option to drop back to the smallest
> common arch during compilation. Note that this is not the same as choosing
> i386 as the target system, since this kernel still has C3 features. Perhaps
> the simplest and most portable way would be to create a new config entry
> which would override default cflags ?
Ok, I added the ability to override the arch-default CFLAGS, but sadly
I'm unable to reproduce your space savings here with gcc 3.3.2. The
default -march=i586 seems to produce the smallest code for me.
tiny-mpm/arch/i386/Makefile | 4 ++++
tiny-mpm/init/Kconfig | 14 ++++++++++++++
2 files changed, 18 insertions(+)
diff -puN arch/i386/Makefile~tiny-cflags arch/i386/Makefile
--- tiny/arch/i386/Makefile~tiny-cflags 2003-12-31 19:02:54.000000000 -0600
+++ tiny-mpm/arch/i386/Makefile 2003-12-31 19:41:14.000000000 -0600
@@ -26,6 +26,9 @@ CFLAGS += $(call check_gcc,-mpreferred-s
align := $(subst -functions=0,,$(call check_gcc,-falign-functions=0,-malign-functions=0))
+ifdef CONFIG_TINY_CFLAGS
+cflags-y += $(CONFIG_TINY_CFLAGS_VAL)
+else
cflags-$(CONFIG_M386) += -march=i386
cflags-$(CONFIG_M486) += -march=i486
cflags-$(CONFIG_M586) += -march=i586
@@ -46,6 +49,7 @@ cflags-$(CONFIG_MWINCHIP2) += $(call che
cflags-$(CONFIG_MWINCHIP3D) += $(call check_gcc,-march=winchip2,-march=i586)
cflags-$(CONFIG_MCYRIXIII) += $(call check_gcc,-march=c3,-march=i486) $(align)-functions=0 $(align)-jumps=0 $(align)-loops=0
cflags-$(CONFIG_MVIAC3_2) += $(call check_gcc,-march=c3-2,-march=i686)
+endif
CFLAGS += $(cflags-y)
diff -puN init/Kconfig~tiny-cflags init/Kconfig
--- tiny/init/Kconfig~tiny-cflags 2003-12-31 19:02:54.000000000 -0600
+++ tiny-mpm/init/Kconfig 2003-12-31 19:44:42.000000000 -0600
@@ -446,6 +446,20 @@ config OOL_THREADINFO
This simplifies the code for finding information about the current
thread. Saves about 4k on small kernels.
+config TINY_CFLAGS
+ default n
+ bool "Set compiler arch flags for small 386 code" if EMBEDDED
+ help
+ This allows user to replace the architecture CFLAGS despite the arch
+ setting in an attempt to build smaller code.
+
+config TINY_CFLAGS_VAL
+ depends TINY_CFLAGS
+ default "-march=i386"
+ string "Arch CFLAGS"
+ help
+ Enter CFLAGS to override ARCH defaults.
+
endmenu # General setup
config SLAB
_
--
Matt Mackall : http://www.selenic.com : Linux development and consulting
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [ANNOUNCE] 2.6.0-tiny1 tree for small systems
2004-01-01 1:46 ` Matt Mackall
@ 2004-01-03 16:50 ` Willy TARREAU
2004-01-03 19:11 ` Matt Mackall
0 siblings, 1 reply; 7+ messages in thread
From: Willy TARREAU @ 2004-01-03 16:50 UTC (permalink / raw)
To: Matt Mackall; +Cc: Willy Tarreau, linux-kernel
Hi Matt,
On Wed, Dec 31, 2003 at 07:46:55PM -0600, Matt Mackall wrote:
> Ok, I added the ability to override the arch-default CFLAGS, but sadly
> I'm unable to reproduce your space savings here with gcc 3.3.2. The
> default -march=i586 seems to produce the smallest code for me.
Same results here. I was a bit stumped at first, but finally found the reason :
by changing CFLAGS, both you and me have implictly removed '-falign-functions=0',
which is what grows up the kernel image ! I really don't understand what happens
in the kernel, because on many other programs, i386 gives me smaller progs than
i586 for -Os optimization, and -falign-functions=0 is *always* smaller. In the
kernel, i386, i586 an c3 are identical, but -falign-functions=0 is bigger by
50 kB :
-Os -mpreferred-stack-boundary=2 -march=c3 -falign-functions=0 -falign-loops=0 -falign-jumps=0 :
text data bss dec hex filename
1306682 111227 50896 1468805 166985 vmlinux
-Os -mpreferred-stack-boundary=2 -march=c3 -falign-loops=0 -falign-jumps=0 :
text data bss dec hex filename
1256604 111227 50896 1418727 15a5e7 vmlinux
I have tested several values and found that the second compilation above is equivalent
to the default gcc setting of -falign-functions=1. And using 1 instead of 0 in
-falign-{loops,jumps,labels} makes not difference at all. So I would suggest
adding -falign-functions=1 everywhere.
BTW, I slightly changed your patch to be able to specify several options. I did
it the dirty way because I don't know how to split the string into several words :
--- ./arch/i386/Makefile Sat Jan 3 15:10:26 2004
+++ a./rch/i386/Makefile Sat Jan 3 15:43:12 2004
@@ -27,7 +27,7 @@
align := $(subst -functions=0,,$(call check_gcc,-falign-functions=0,-malign-functions=0))
ifdef CONFIG_TINY_CFLAGS
-cflags-y += $(CONFIG_TINY_CFLAGS_VAL)
+cflags-y += $(shell echo $(CONFIG_TINY_CFLAGS_VAL))
else
cflags-$(CONFIG_M386) += -march=i386
cflags-$(CONFIG_M486) += -march=i486
Other than that, I've compiled 2.6.1-rc1-tiny1. It's slightly smaller than 2.6.0-tiny1
here. But I had to keep CONFIG_INETPEER=y, CONFIG_DNOTIFY=y, CONFIG_PTRACE=y,
and CONFIG_CPU_SUP_* otherwise it would not link. I can give you the error reports
and .config in case you're interested.
Cheers,
Willy
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [ANNOUNCE] 2.6.0-tiny1 tree for small systems
2004-01-03 16:50 ` Willy TARREAU
@ 2004-01-03 19:11 ` Matt Mackall
0 siblings, 0 replies; 7+ messages in thread
From: Matt Mackall @ 2004-01-03 19:11 UTC (permalink / raw)
To: Willy TARREAU; +Cc: linux-kernel
On Sat, Jan 03, 2004 at 05:50:51PM +0100, Willy TARREAU wrote:
> BTW, I slightly changed your patch to be able to specify several
> options. I did it the dirty way because I don't know how to split
> the string into several words :
Doh, I fixed that but it didn't make it into the release:
+ifdef CONFIG_TINY_CFLAGS
+cflags-y += $(shell echo $(CONFIG_TINY_CFLAGS_VAL) | sed -e 's/"//g')
+else
> Other than that, I've compiled 2.6.1-rc1-tiny1. It's slightly
> smaller than 2.6.0-tiny1 here. But I had to keep CONFIG_INETPEER=y,
> CONFIG_DNOTIFY=y, CONFIG_PTRACE=y, and CONFIG_CPU_SUP_* otherwise it
> would not link. I can give you the error reports and .config in case
> you're interested.
Hmmm, works for me, error messages appreciated.
--
Matt Mackall : http://www.selenic.com : Linux development and consulting
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2004-01-03 19:11 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2003-12-27 21:56 [ANNOUNCE] 2.6.0-tiny1 tree for small systems Matt Mackall
2003-12-28 10:35 ` Willy Tarreau
2004-01-01 1:46 ` Matt Mackall
2004-01-03 16:50 ` Willy TARREAU
2004-01-03 19:11 ` Matt Mackall
2003-12-28 23:54 ` Mike Fedyk
2003-12-29 0:31 ` William Lee Irwin III
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox