Hi Alexander,
Alexander Eremeenkov wrote:
I have performance problems with applications on this kernel(uClinux
2.6.25-uc0).
First of all, I have custom made board with this features:
ColdFire 5274 @ 150 MHz
External Bus Frequency 75 MHz
uClinux 2.6.25 boot up and work perfectly stable, but very slow. For
this board i have also 2.4.x kernel compiled, and it works more faster (
~ 2-20 times faster, if different test application).
After reading maillist, i find, that some people have equals problems,
and their reason were - disabled cache. But in my case, cache enabled in
start process normally.
Calibration delay calculate good value:
Calibrating delay loop... 98.71 BogoMIPS (lpj=493568)
Also, if i disabled cache in /include/asm-m68knommu/mcfcache.h, i got
looks-like normal, with disabled cache, value:
Calibrating delay loop... 5.82 BogoMIPS
So, I have drawn a conclusion, that cache enables good.
Now, about performance. I test it with some applications.
1. Dhrystone test. With it, i have 4065 dhrystones/second, that
critically small for this CPU. But i curious, that, when a disabled
cache, the result same.. As though like cache at just flushed or worked
incorrect.
2. Simple forever cycle with 1 value in memory incremented. This test,
also show near ~5 real MISP performance.
When i try to watch assess to external memory by CPU, i saw, that, it
work with external RAM(where this application contain) everytime - but i
think that, simple application with 5 command + 1 value memory must
being in cache, and CPU must work with cache only.
3. Some real/work application. They do massive processing/moving data,
working with peripheral, etc. I also have quite bad performance. Those
applications runs at 2.4.x kernel at ~4 times faster.
So, conclusion with it.
1) I think, that i can't have problems with hardware - because, 2.4.x
kernel work fast.
2) I have enabled cache at start process - 98 BogoMIPS good value for my
CPU, also 2.4 kernel calculate same result.
Now i have question, maybe somebody have same problems at this kernel?
Maybe 2.6.xx so slow, and ways to speed up it use old kernel or
modern/faster CPU?Maybe some bugs in poring it to ColdFire family?Etc?
There is no reason that application code performance (independant of
use of kernel system calls) should give noticably different results
on a 2.6 kernel vs a 2.4 kernel.
Maybe it problem with cache at working process? Some drivers, for
example, can flush cache..anyone have equal problem? About it, i will
check it with kernel build for 5275EVB, but i think, that result will
be same, because, a removed all i can drivers/modules form kernel, that
almost "empty" kernel starts, and after it i run Dhrystone test, and get
same result.
Maybe it's a toolchain problem? I use m68k-uclinux-tools-20061214.sh,
they have some minor problems, maybe reason with they? But, my second
test with simple application excludes this variant.
If someone had those problems I will be glad for any help.
I recall a problem a little while back where the cache flush code
was changing the cache configuration and not just flushing. (I think
it was the 5282 cache support code that was broken). In this scenario
the initial cache setup was good (so everything was fast), and after
the first cache flush the setup was wrong. Now that would make something
like the bogomips calculation look good, but later performance bad
Looking at the 2 places this is done:
linux-2.6.x/include/asm-m68knommu/mcfcache.h
linux-2.6.x/include/asm-m68knommu/cacheflush.h
I suspect this may have broken the cache support for the 527x
series (so the 5270/5271 and 5274/5275).
To verify if this is what you are seeing, can you change the cache
flush code for CONFIG_M527x in cacheflush.h from:
"movel #0x81000200, %%d0\n\t"
to
"movel #0x81400100, %%d0\n\t"
This is just to prove this is the problem. A real fix would
need the 528x and 527x cache flushing code separated out.
Regards
Greg
And my kernel messages at result:
/> cat
/proc/kmsg
<5>Linux version 2.6.25-uc0 (wa...@arch) (gcc version 4.1.1) #36 Mon Jan
5 13:05:16 PST 9
<6>
<4>
<4>uClinux/COLDFIRE(m5274/5275)
<6>COLDFIRE port done by Greg Ungerer,
g...@snapgear.com <6>Flat model support
(C) 1998,1999 Kenneth Albanowski, D. Jeff Dionne <7>On
node 0 totalpages:
8192 <7> DMA
zone: 0 pages used for
memmap <7> Normal zone:
64 pages used for memmap
<7> Normal zone: 8128 pages, LIFO
batch:0 <7> Movable zone: 0
pages used for memmap
<4>Built 1 zonelists in Zone order, mobility grouping on. Total pages:
8128 <5>Kernel command
line:
<6>Dentry cache hash table entries: 4096 (order:
2, 16384 bytes) <6>Inode-cache hash table entries:
2048 (order: 1, 8192 bytes) <6>Memory available:
29264k/32768k RAM, (1504k kernel code, 243k data)
<7>Calibrating delay loop... 98.71 BogoMIPS
(lpj=493568) <4>Mount-cache hash table
entries: 512
<6>net_namespace: 144
bytes <6>NET:
Registered protocol family
16 <6>NET: Registered
protocol family 2
<6>IP route cache hash table entries: 1024 (order: 0, 4096
bytes) <6>TCP established hash table entries: 1024
(order: 1, 8192 bytes) <6>TCP bind hash table
entries: 1024 (order: 0, 4096 bytes) <6>TCP:
Hash tables configured (established 1024 bind
1024) <6>TCP reno
registered
<6>Installing knfsd (copyright (C) 1996
o...@monad.swb.de). <6>JFFS2 version 2.2.
(NAND) ?? 2001-2006 Red Hat, Inc. <6>io
scheduler noop
registered <6>io
scheduler cfq registered
(default)
<4>ColdFire internal UART serial
driver <6>ttyS0 at MMIO
0x40000200 (irq = 77) is a ColdFire UART
<6>console [ttyS0]
enabled
<6>ttyS1 at MMIO 0x40000240 (irq = 78) is a ColdFire
UART <6>ttyS2 at MMIO 0x40000280 (irq = 79)
is a ColdFire UART <6>brd: module
loaded
<4>FEC ENET Version
0.2
<4>fec: PHY @ 0x1, ID 0x20005c90 --
DP83848 <4>eth0: ethernet
00:04:24:23:34:44
<4>uclinux[mtd]: RAM probe address=0x1d50f4
size=0x14d000 <5>Creating 1 MTD partitions
on "RAM":
<5>0x00000000-0x0014d000 :
"ROMfs"
<4>uclinux[mtd]: set ROMfs to be root
filesystem
<6>Physically mapped flash: Found 1 x16 devices at 0x0 in 16-bit
bank <4> Intel/Sharp Extended Query Table at
0x010A <4> Intel/Sharp Extended
Query Table at 0x010A <4>
Intel/Sharp Extended Query Table at
0x010A <4> Intel/Sharp Extended
Query Table at 0x010A <4>
Intel/Sharp Extended Query Table at
0x010A <6>Using buffer write
method
<6>Using auto-unlock on
power-up/resume
<5>cfi_cmdset_0001: Erase suspend on write
enabled <7>erase region 0:
offset=0x0,size=0x8000,blocks=4
<7>erase region 1:
offset=0x20000,size=0x20000,blocks=63
<6>TCP cubic
registered
<6>NET: Registered protocol family
1 <6>RPC: Registered
udp transport module.
<6>RPC: Registered tcp transport
module. <6>NET: Registered
protocol family 33
<4>VFS: Mounted root (romfs filesystem)
readonly. <5>Freeing unused kernel
memory: 52k freed (0x1a7000 - 0x1b3000) <4>eth0:
config: auto-negotiation on, 100FDX, 100HDX, 10FDX,
10HDX.
--
------------------------------------------------------------------------
Greg Ungerer -- Principal Engineer EMAIL: g...@snapgear.com
SnapGear, a McAfee Company PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com
_______________________________________________
uClinux-dev mailing list
uClinux-dev@uclinux.org
http://mailman.uclinux.org/mailman/listinfo/uclinux-dev
This message was resent by uclinux-dev@uclinux.org
To unsubscribe see:
http://mailman.uclinux.org/mailman/options/uclinux-dev
