Performance Issue on Coldfire M548x.

[Groleo]

Hi,

I'm doing some speed tests on a M548x and it seems a bit sluggish.

To begin, the OS is a linux-2.6.25.20 patched using the attached list of patches.

The patches we're downloaded from www.bitshrine.org while The compiler is a

vanilla gcc-4.4.0, without the CLoop and PPL included.

cat /proc/cpuinfo
CPU:            ColdFire V4e
MMU:            ColdFire
FPU:            ColdFire V4e
Clocking:       198.2MHz
BogoMips:       198.24
Calibration:    991232 loops

Now, onto the tests.

I ran simple file copy from/to flash  to/from RAM as can be seen

in the attached bench.sh.

/tmp is a RAM disk and / is the flash.

This is what I got from running bench.sh on ColdFire

time cat /tmp/test_file > /dev/null
real    0m 0.09s
user    0m 0.00s
sys     0m 0.09s
time cat /tmp/test_file > /test_file
real    0m 17.81s
user    0m 0.01s
sys     0m 17.80s
time cat /test_file > /dev/null
real    0m 0.10s
user    0m 0.01s
sys     0m 0.09s
time cat /test_file > /tmp/test_file2
real    0m 0.28s
user    0m 0.02s
sys     0m 0.27s
time cat /tmp/test_file > /tmp/test_file2
real    0m 0.29s
user    0m 0.00s
sys     0m 0.29s
/tmp

The same test  was performed on an ARM

cat /proc/cpuinfo
Processor       : ARM920T rev 0 (v4l)
BogoMIPS        : 99.73
Features        : swp half thumb crunch

With the following results:

time cat /tmp/test_file > /dev/null
real    0m 0.04s
user    0m 0.00s
sys     0m 0.04s
time cat /tmp/test_file > /test_file
real    0m 11.28s
user    0m 0.00s
sys     0m 7.15s
time cat /test_file > /dev/null
real    0m 0.04s
user    0m 0.00s
sys     0m 0.03s
time cat /test_file > /tmp/test_file2
real    0m 0.07s
user    0m 0.00s
sys     0m 0.05s
time cat /tmp/test_file > /tmp/test_file2
real    0m 0.06s
user    0m 0.00s
sys     0m 0.06s
/tmp

You can see that the test `time cat /tmp/test_file > /tmp/test_file2` performs real bad on

the ColdFire.

Have you any clues on this one ?

[Groleo]

To make things clear, I placed the values in a table.

You can see that there is a problem when writing to Memory

 CuCu | Z->M | M->N | M->M |  M->F | F->N |  F->F | F->M |
------+------+------+------+-------+------+-------+------+    
VR900 | 0.24 | 0.08 | 0.27 | 17.82 | 0.08 | 17.73 | 0.27 |
ARM   | 0.06 | 0.04 | 0.07 | 09.56 | 0.04 |  9.57 | 0.07 |
VR/ARM|   4  |   2  | 3.85 |  1.86 |   2  |  1.8  | 3.85 |

Z->M : read from /dev/zero, write to Memory.
M->N : read from Memory, write to /dev/null
M->M : Read from Memory, write to Memory
M->F : read from Memory, write to Flash
F->N : read from Flash, write to /dev/null
F->M : read from Flash, write to Memory

[meteorcool]

 Could you post your memory settings in include/asm-m68k/cf_pgtable.h arch/m68k/coldfire/head.S and the assembly code for the memory test?  

[Groleo]

I attached the requested files.

For memory test I used

dd if=/dev/zero of=/tmp/test_file bs=1M count=1, with the "dd" from busybox-1.13.4[1]

[1]: http://git.busybox.net/busybox/tree/coreutils/dd.c?id=1_13_4

This is the strace output:

# strace dd if=/dev/zero of=/tmp/test_file bs=1M count=1
execve("/bin/dd", ["dd", "if=/dev/zero", "of=/tmp/test_file", "bs=1M", "count=1"], [/* 14 vars */]) = 0
old_mmap(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0x75) = 0x8006a000
stat("/etc/ld.so.cache", 0xbfac9998)    = -1 ENOENT (No such file or directory)
stat("/etc/ld.so.preload", 0xbfac9b34)  = -1 ENOENT (No such file or directory)
open("/lib/libc.so.0", O_RDONLY)        = 3
fstat(3, {st_mode=S_IFREG|0444, st_size=233724, ...}) = 0
old_mmap(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0x75) = 0x8006c000
read(3, "\177ELF\1\2\1\0\0\0\0\0\0\0\0\0\0\3\0\4\0\0\0\1\0\0\270|\0\0\0004\0"..., 8192) = 8192
old_mmap(NULL, 262144, PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0x75) = 0x8006e000
old_mmap(0x8006e000, 229924, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_FIXED, 3, 0x75) = 0x8006e000
old_mmap(0x800a8000, 4348, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED, 3, 0x3800000000075) = 0x800a8000
old_mmap(0x800aa000, 14916, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0x75) = 0x800aa000
close(3)                                = 0
munmap(0x8006c000, 8192)                = 0
stat("/lib/ld-uClibc.so.0", {st_mode=S_IFREG|0555, st_size=15780, ...}) = 0
ioctl(0, SNDCTL_TMR_TIMEBASE or TCGETS, {B115200 opost isig icanon echo ...}) = 0
ioctl(1, SNDCTL_TMR_TIMEBASE or TCGETS, {B115200 opost isig icanon echo ...}) = 0
getuid32()                              = 0
old_mmap(NULL, 1056768, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, 0, 0x75) = 0x800ae000
open("/dev/zero", O_RDONLY)             = 3
dup2(3, 0)                              = 0
close(3)                                = 0
open("/tmp/test_file", O_WRONLY|O_CREAT|O_TRUNC, 0666) = 3
dup2(3, 1)                              = 1
close(3)                                = 0
read(0, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 1048576) = 1048576
write(1, "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 1048576) = 1048576
close(0)                                = 0
close(1)                                = 0
write(2, "1"..., 11)                     = 1
write(2, "+"..., 1+)                     = 1
write(2, "0"..., 10)                     = 1
write(2, " records in\n"..., 12 records in
)        = 12
write(2, "1"..., 11)                     = 1
write(2, "+"..., 1+)                     = 1
write(2, "0"..., 10)                     = 1
write(2, " records out\n"..., 13 records out
)       = 13
_exit(0)                                = ?