[#5874] malloc performance frequently fails on bf561 SMP kernel

Submitted By: Mingquan Pan

Open Date

2010-01-28 23:18:07 Close Date

2010-02-11 03:51:10

Priority:

Medium Assignee:

Graf Yang

Status:

Closed Fixed In Release:

N/A

Found In Release:

2010R1 Release:

Category:

N/A Board:

N/A

Processor:

BF561 Silicon Revision:

Is this bug repeatable?:

Yes Resolution:

Rejected

Uboot version or rev.:

Toolchain version or rev.:

4.3.4 (ADI-trunk/svn-3771)

App binary format:

N/A

Summary: malloc performance frequently fails on bf561 SMP kernel

Details:

malloc performance frequently fails on bf561 SMP kernel now.

Linux version 2.6.32.2-ADI-2010R1-pre-svn8124 (test@uclinux65-561-SMP) (gcc version 4.3.4 (ADI-trunk/svn-3771) ) #40 SMP Thu Jan 7 11:54:37 GMT 2010

bootconsole [early_shadow0] enabled

bootconsole [early_BFuart0] enabled

early printk enabled on early_BFuart0

Board Memory: 64MB

Kernel Managed Memory: 64MB

Memory map:

fixedcode = 0x00000400-0x00000490

text = 0x00001000-0x0010c010

rodata = 0x0010c020-0x0015ebf0

bss = 0x0015f000-0x001714c8

data = 0x001714e0-0x00182000

stack = 0x00180000-0x00182000

init = 0x00182000-0x006d6000

available = 0x006d6000-0x03f00000

DMA Zone = 0x03f00000-0x04000000

Hardware Trace Active and Enabled

Boot Mode: 0

Reset caused by Software reset

Compiled for ADSP-BF561 Rev 0.5

Blackfin Linux support by http://blackfin.uclinux.org/

Processor Speed: 600 MHz core clock and 100 MHz System Clock

NOMPU: setting up cplb tables

Instruction Cache Enabled for CPU0

External memory: cacheable in instruction cache

L2 SRAM : uncacheable in instruction cache

Data Cache Enabled for CPU0

External memory: cacheable (write-through) in data cache

L2 SRAM : uncacheable in data cache

Built 1 zonelists in Zone order, mobility grouping off. Total pages: 16002

Kernel command line: root=/dev/mtdblock0 rw ip=10.100.4.50 earlyprintk=serial,uart0,57600 console=ttyBF0,57600 ip=10.100.4.50:10.100.4.174:10.100.4.174:255.255.255.0:bf561-ezkit:eth0:off

PID hash table entries: 256 (order: -2, 1024 bytes)

Dentry cache hash table entries: 8192 (order: 3, 32768 bytes)

Inode-cache hash table entries: 4096 (order: 2, 16384 bytes)

Memory available: 56916k/65536k RAM, (5456k init code, 1068k kernel code, 472k data, 1024k dma, 600k reserved)

Hierarchical RCU implementation.

NR_IRQS:121

Configuring Blackfin Priority Driven Interrupts

console [ttyBF0] enabled, bootconsole disabled

Calibrating delay loop... 1187.84 BogoMIPS (lpj=2375680)

Mount-cache hash table entries: 512

CoreB bootstrap code to SRAM ff600000 via DMA.

Booting Core B.

Instruction Cache Enabled for CPU1

External memory: cacheable in instruction cache

L2 SRAM : uncacheable in instruction cache

Data Cache Enabled for CPU1

External memory: cacheable (write-through) in data cache

L2 SRAM : uncacheable in data cache

Brought up 2 CPUs

Calibrating delay loop...

SMP: Total of 2 processors activated (4.09 BogoMIPS).

Blackfin Scratchpad data SRAM: 4 KB

Blackfin L1 Data A SRAM: 16 KB (16 KB free)

Blackfin L1 Data B SRAM: 16 KB (16 KB free)

Blackfin L1 Instruction SRAM: 16 KB (15 KB free)

Blackfin L2 SRAM: 128 KB (127 KB free)

NET: Registered protocol family 16

Blackfin DMA Controller

ezkit_init(): registering device resources

bio: create slab <bio-0> at 0

Switching to clocksource jiffies

NET: Registered protocol family 2

IP route cache hash table entries: 1024 (order: 0, 4096 bytes)

TCP established hash table entries: 2048 (order: 2, 16384 bytes)

TCP bind hash table entries: 2048 (order: 2, 16384 bytes)

TCP: Hash tables configured (established 2048 bind 2048)

2138.11 BogoMIPS (lpj=4276224)

TCP reno registered

NET: Registered protocol family 1

msgmni has been set to 111

io scheduler noop registered

io scheduler anticipatory registered (default)

bfin-uart: Blackfin serial driver

bfin-uart.0: ttyBF0 at MMIO 0xffc00400 (irq = 35) is a BFIN-UART

brd: module loaded

bfin-spi bfin-spi.0: Blackfin on-chip SPI Controller Driver, Version 1.0, regs_base@ffc00500, dma channel@16

smc91x.c: v1.1, sep 22 2004 by Nicolas Pitre <nico@fluxnic.net>

eth0: SMC91C11xFD (rev 2) at 2c010300 IRQ 82 [nowait]

eth0: Ethernet addr: 00:e0:22:fe:ba:2a

bfin-wdt: initialized: timeout=20 sec (nowayout=0)

TCP cubic registered

NET: Registered protocol family 17

eth0: link down

IP-Config: Complete:

device=eth0, addr=10.100.4.50, mask=255.255.255.0, gw=10.100.4.174,

host=bf561-ezkit, domain=, nis-domain=(none),

bootserver=10.100.4.174, rootserver=10.100.4.174, rootpath=

Freeing unused kernel memory: 5456k freed

dma_alloc_init: dma_page @ 0x02786000 - 256 pages at 0x03f00000

eth0: link up, 100Mbps, full-duplex, lpa 0x41E1

_____________________________________

a8888b. / Welcome to the uClinux distribution \

d888888b. / _ _ \

8P"YP"Y88 / | | |_| __ __ (TM) |

8|o||o|88 _____/ | | _ ____ _ _ \ \/ / |

8' .88 \ | | | | _ \| | | | \ / |

8`._.' Y8. \ | |__ | | | | | |_| | / \ |

d/ `8b. \ \____||_|_| |_|\____|/_/\_\ |

dP . Y8b. \ For embedded processors including |

d8:' " `::88b \ the Analog Devices Blackfin /

d8" 'Y88b \___________________________________/

:8P ' :888

8a. : _a88P For further information, check out:

._/"Yaa_: .| 88P| - http://blackfin.uclinux.org/

\ YP" `| 8P `. - http://docs.blackfin.uclinux.org/

/ \.___.d| .' - http://www.uclinux.org/

`--..__)8888P`._.' jgs/a:f - http://www.analog.com/blackfin

Have a lot of fun...

BusyBox v1.15.3 (2010-01-07 04:40:08 GMT) hush - the humble shell

root:/> version

kernel: Linux release 2.6.32.2-ADI-2010R1-pre-svn8124, build #40 SMP Thu Jan 7 11:54:37 GMT 2010

toolchain: bfin-uclinux-gcc release gcc version 4.3.4 (ADI-trunk/svn-3771)

user-dist: release svn-9347, build #461 Thu Jan 7 11:53:29 GMT 2010

root:/> successful boot attempt

************** STEP 3: Start testing.

uname -a

Linux blackfin 2.6.32.2-ADI-2010R1-pre-svn8124 #40 SMP Thu Jan 7 11:54:37 GMT 2010 blackfin GNU/Linux

root:/> malloc-perf 120

00004k : 0x02772004 000000 000000 000000

00008k : 0x020a0004 000000 000000 000000

00012k : 0x020b4004 000000 000031 004000

00016k : 0x02948004 000000 000000 000000

00020k : 0x02950004 000000 000000 000000

00024k : 0x02958004 000000 000000 000000

00028k : 0x02960004 000000 000000 000000

00032k : 0x02968004 000000 000000 000000

00036k : 0x02970004 000000 000062 004000

00040k : 0x02970004 000000 000093 004000

00044k : 0x02970004 000000 000031 004000

00048k : 0x02970004 000000 000031 004000

00052k : 0x02970004 000000 000031 004000

00056k : 0x02970004 000000 000031 004000

00060k : 0x02970004 000000 000031 004000

00064k : 0x02970004 000000 000000 000000

00068k : 0x02a00004 000000 000062 004000

00072k : 0x02a00004 000000 000062 004000

00076k : 0x02a00004 000000 000062 004000

00080k : 0x02a00004 000000 000031 004000

00084k : 0x02a00004 000000 000031 004000

00088k : 0x02a00004 000000 000062 004000

00092k : 0x02a00004 000000 000031 004000

00096k : 0x02a00004 000000 000031 004000

00100k : 0x02a00004 000000 000031 004000

00104k : 0x02a00004 000000 000062 004000

00108k : 0x02a00004 000000 000062 004000

00112k : 0x02a00004 000000 000031 004000

00116k : 0x02a00004 000000 000062 004000

00120k : 0x02a00004 000000 000062 004000

00124k : 0x02a00004 000000 000031 004000

00128k : 0x02a00004 000000 000031 004000

00256k : 0x02a00004 000000 000031 004000

00384k : 0x02a00004 000000 000125 004000

00512k : 0x02a00004 000000 000031 004000

00640k : 0x02a00004 000000 000125 004000

00768k : 0x02a00004 000000 000156 004000

00896k : 0x02a00004 000000 000125 004000

01024k : 0x02a00004 000000 000156 004000

TEST FAIL

Follow-ups

--- Graf Yang 2010-01-30 07:47:22

The SMP kernel performance is a bit lower than the UP kernel. If it need 20%-30%

more time to finish the malloc performance test, I think it is normal.

Can you point out a probable time that the fail begins frequently?

--- Mingquan Pan 2010-02-01 04:29:00

The log of Nov 27 has much better result, it can pass 9 of 10 times.

root:/> version

kernel: Linux release 2.6.31.6-ADI-2010R1-pre-svn7883, build #105 SMP Fri

Nov 27 09:24:48 GMT 2009

toolchain: bfin-uclinux-gcc release gcc version 4.3.4 (ADI-trunk/svn-3679)

user-dist: release svn-9185, build #1192 Fri Nov 27 09:23:32 GMT 2009

root:/> successful boot attempt

************** STEP 3: Start testing.

root:/> malloc-perf 120

00004k : 0x02bfb004 000000 000000 000000

00008k : 0x02b6e004 000000 000000 000000

00012k : 0x02bec004 000000 000000 000000

00016k : 0x020cc004 000000 000000 000000

00020k : 0x02ae8004 000000 000000 000000

00024k : 0x02a68004 000000 000000 000000

00028k : 0x02960004 000000 000000 000000

00032k : 0x029e8004 000000 000031 004000

00036k : 0x02a70004 000000 000031 004000

00040k : 0x02a70004 000000 000000 000000

00044k : 0x02a70004 000000 000062 004000

00048k : 0x02a70004 000000 000062 004000

00052k : 0x02a70004 000000 000062 004000

00056k : 0x02a70004 000000 000031 004000

00060k : 0x02a70004 000000 000000 000000

00064k : 0x02a70004 000000 000031 004000

00068k : 0x02840004 000000 000031 004000

00072k : 0x02840004 000000 000062 004000

00076k : 0x02840004 000000 000062 004000

00080k : 0x02840004 000000 000062 004000

00084k : 0x02840004 000000 000031 004000

00088k : 0x02840004 000000 000031 004000

00092k : 0x02840004 000000 000000 000000

00096k : 0x02840004 000000 000093 004000

00100k : 0x02840004 000000 000093 004000

00104k : 0x02840004 000000 000031 004000

00108k : 0x02840004 000000 000062 004000

00112k : 0x02840004 000000 000093 004000

00116k : 0x02840004 000000 000093 004000

00120k : 0x02840004 000000 000000 000000

00124k : 0x02840004 000000 000000 000000

00128k : 0x02840004 000000 000000 000000

00256k : 0x02800004 000000 000062 004000

00384k : 0x02c00004 000000 000000 000000

00512k : 0x02c00004 000000 000125 004000

00640k : 0x02c00004 000000 000187 004000

00768k : 0x02c00004 000000 000000 000000

00896k : 0x02c00004 000000 000062 004000

01024k : 0x02c00004 000000 000000 000000

TEST PASS

--- Graf Yang 2010-02-01 22:43:23

I found the new SMP kernel has lower malloc performance over the old(2009R1 for

example). I'd dig out the reason.

--- Graf Yang 2010-02-03 06:26:16

Not bug.

Since kernel updated to 2.6.32, the function prep_new_page(struct page *page,

int order, gfp_t gfp_flags) will check every page, other than only check the

first page. This makes the malloc-perf will take double time when malloc

memory.

So I suggest double the threshold to 200,

malloc-perf 200

BTW, this test need enable at least one CLOCKSOURCE options.

--- Graf Yang 2010-02-03 21:16:13

Current SMP kernel 180, UP 90 should pass the test.

--- Mingquan Pan 2010-02-11 03:50:33

After increasing the param, it can pass now.

root:/> malloc-perf 180^M

00004k : 0x02775004 000003 000004 000046^M

00008k : 0x027ea004 000003 000004 000040^M

00012k : 0x0289c004 000003 000004 000039^M

00016k : 0x027ea004 000003 000004 000046^M

00020k : 0x027ea004 000003 000003 000006^M

00024k : 0x027ea004 000003 000004 000030^M

00028k : 0x028c0004 000004 000004 000054^M

00032k : 0x028c8004 000031 000033 000086^M

00036k : 0x028d0004 000034 000037 000105^M

00040k : 0x028d0004 000034 000037 000097^M

00044k : 0x028d0004 000035 000037 000070^M

00048k : 0x028d0004 000034 000038 000108^M

00052k : 0x028d0004 000034 000046 001117^M

00056k : 0x028d0004 000034 000037 000103^M

00060k : 0x028d0004 000035 000037 000092^M

00064k : 0x028d0004 000035 000037 000106^M

00068k : 0x028e0004 000041 000044 000104^M

00072k : 0x028e0004 000041 000045 000193^M

00076k : 0x028e0004 000041 000046 000110^M

00080k : 0x028e0004 000041 000045 000111^M

00084k : 0x028e0004 000041 000044 000113^M

00088k : 0x028e0004 000041 000044 000101^M

00092k : 0x028e0004 000040 000044 000102^M

00096k : 0x028e0004 000040 000043 000113^M

00100k : 0x028e0004 000041 000045 000115^M

00104k : 0x028e0004 000041 000044 000098^M

00108k : 0x028e0004 000041 000045 000118^M

00112k : 0x028e0004 000041 000045 000110^M

00116k : 0x028e0004 000041 000045 000112^M

00120k : 0x028e0004 000041 000045 000138^M

00124k : 0x028e0004 000041 000045 000103^M

00128k : 0x028e0004 000041 000045 000104^M

00256k : 0x02a00004 000055 000068 001129^M

00384k : 0x02a00004 000081 000091 000164^M

00512k : 0x02a00004 000081 000091 000169^M

00640k : 0x02a00004 000134 000145 000206^M

00768k : 0x02a00004 000134 000148 000214^M

00896k : 0x02a00004 000134 000150 000243^M

01024k : 0x02a00004 000134 000149 000276^M

TEST PASS^M

root:/> malloc-perf pass

Files

Changes

Commits

Dependencies

Duplicates

Associations

Partner Forums

Partner Libraries

[#5874] malloc performance frequently fails on bf561 SMP kernel