重點在 namespaceId 的值，它是 hash( host+port )。
如果其他的 namenode 不合會怎樣？
不能啟動嗎？
總不可能每次 connect 都檢查這個值吧？

just do it!!!

TODO:
namenode namesystem.blockReceived

整理 code ？
1. prologue 及 epilogue。
2. indirect branch -> direct brance
3. code cache memory manager
4.

build hadoop

2010/06/03 build hadoop, problem:
ivy-download:
/home/tk/Software/hadoop/hadoop-svn-0.20.2/build.xml:1630: java.net.SocketException: Network is unreachable
mentioned in this post that :

Setting bindv6only to 0 in /etc/sysctl.d/bindv6only.conf on my debian squeeze installation seems to have fixed the problem. Sorry for littering the list.

This actually is a special issue in Debina squeeze/sid version as explained in another post.
So what to do is:
1. Setting bindv6only to 0 in /etc/sysctl.d/bindv6only.conf
2. restart procps: sudo invoke-rc.d procps restart

now we have problem for java version

debug

目前想改 repz 的instruction，讓它們跟 qemu 的行為一樣，這樣比較好debug，這很重要，雖然implement 上會比較慢，但開發速度可以快很多，確定都ok了之後就可以改回來。

what's the major challenge of retargetable binary translator
I think the binary translator currently does not has enough tool to run optimization

what's the benefit to convert binary code to high level intermediate code?

now we have LLVM tool to manipulate binary at runtime.
to map guest register to host register at runtime.

gogogogogogogogogogogogogogogogogo

do I need to start writing this paper?

tobuy:
床架、布？以後也許不太會在客廳也不一定
桌子要擺飯廳
要

今天要完成其他的app，要先列出順序

floating point 有點麻煩，要自己寫，用 qemu 處理 exception
qemu 有處理 pre-op float exception，但沒有 post-op float exception

記得將 hot path problem 寫成 paper

最近心情有點煩

debug不知如何進行！

又要搬家

一個人住好了，簡單，但花錢！

目前有兩個大煩惱：debug如何進行，以及hadoop的project。
我覺得這兩個都是很有趣的問題，不太算煩惱的煩惱。
搬家雖然很煩，但是個改變，感覺也不錯，也許先搬過去再找房客也不差，不急。

下星期會很忙！就星期五搬好了。冷氣要跟室友談好，

加油！

反覆的開啟又關閉

我不能讓一件事就打亂我的心情，

乾脆，那間會很恐怖嗎？
不知道，不過，

抖抖抖抖抖抖的的的抖抖抖抖

qemu 在 user mode 下是用 static code gen buffer，size為32mb
in ./exec.c line 408 it mentioned:

Currently it is not recommended to allocate big chunks of data in
user mode. It will change when a dedicated libc will be used

but don't know why. Why not use mmap as system mode use?

Segment selector Format:

15 ...... 3	2 1	0
index	TI	RPL

TI Table indicator:

0 means selector indexes into GDT
1 means selector indexes into LDT

RPL Privelege level. Linux uses only two privelege levels.

0 means kernel
3 means user

Examples:

Kernel code segment

TI=0, index=1, RPL=0, therefore selector = 0x08 (GDT[1])

User data segment

TI=1, index=2, RPL=3, therefore selector = 0x17 (LDT[2])

stack 弄壞了！

先弄一個 exit basic block，然後再將 pass 給去掉。

現在要focus 在 indirect branch 上，要知道的是第一個參數的 address 是什麼？
FP->getPassName()
lib/CodeGen/StackProtector.cpp

DefaultJITMemoryManager

lib/CodeGen/LLVMTargetMachine.cpp:
LLVMTargetMachine::addPassesToEmitMachineCode
LLVMTargetMachine::addCommonCodeGenPasses

=============================
Prevent spill register code gen.
enable to emit epilog for branch and indirect brance
=============================
錯在這：0x40095872: mov (%edx),%eax
%edx 為 init_stack+c 的位置，其中的值
在 loader_exec 中就放了！
在 create_elf_tables() 之後！
在 loader_build_argptr() 之後！
在 put_user_ual(stringp, envp) 之後！

MOVS：move string from ds:esi to es:edi
this confuses me

Question is
address-size attribute

給ice：(guest_ip，target_ip，count，type)

what are the meanings of cf,of and af flags in eflag？

CF: This flag indicates an overflow condition for unsigned-integer arithmetic.
Set if an arithmetic operation generates a carry or a borrow out of the most-significant bit of the result; cleared otherwise.

OF: This flag indicates an overflow condition for signed-integer (two’s complement) arithmetic.
Set if the integer result is too large a positive number or too small a negative number (excludingthe sign-bit) to fit in the destination operand; cleared otherwise.

AF: This flag is used in binary-coded decimal (BCD) arithmetic.
Set if an arithmetic operation generates a carry
or a borrow out of bit 3 of the result; cleared otherwise.

首先：

我要將 translation funtion的界面換掉，
它要有：
一、 disassembler傳過去的參數。
二、是否要更新 cc code。
三、extra parameter。

MOV16sr: move register into seqment register
store guest ip into

I summarize our discussion as follows.

As stated in the CACM article, the single master may pose some problems when tremendous amount of data are considered.
For example, the master is unable to store all the metadata in the memory.
In addition, the master would become the bandwidth bottleneck because it has to a large amount of connections.
Google now also wants to deal with this issue, and they propose to use multiple masters.
In their proposal, a file will be related to one of the masters in an static/unchangeable way.

Besides, it seems that they do not consider the load balance problem of masters in their design.

So, the following is our design, in which the load balance and bandwidth problems are simultaneously solved.
Assume that we have k masters, each of which takes care of certain chunkservers.

Each master stores k counting bloom filters, among which i-th bloom filter represents the association between the i-th master and the file-chunk mapping it is in charge of.

Note that each master broadcasts its association to the other masters so that each master can renew the counting bloom filters it stores.

When a client wants to access a file, what it should do is to randomly pick a master and then to query the corresponding k counting bloom filters.

If not hit happens in all the filters, it means that the file does not exist.

Otherwise, the client turns to ask the master that is in charge of the file the client would like to access.

Of course, there could be cases that more than one masters are responsible for the file the client would like to access according to the response of the filters.

This is due to the nature of bloom filter, and can be easily mitigated and handled.

Increasing the filter size can mitigate such rate of false hit.

Simply asking those two masters to make sure again can be helpful in handling this ambiguity.

As a whole, this design can solve the memory problem because k masters share the workload.

This design can solve the bandwidth problem because of the randomness in the choice made by the client.

I think we can raise the following questions:

1. Why the engineers in Google insist that, in the single master case, all the metadata should be stored in the memory in the master?

A: First, storing some of metadata in the disk may slow down the overall access speed. Second, doing so still cannot solve the bandwidth problem.

2. The CACM ariticle briefs Google's plan on the multiple master setting in GFS. Do you have any idea about their design? Besides, Do you have any idea how to coordinate multiple masters?

A: (We can decribe our above idea. I draw a conceptional picture. Perhaps we can attach this picture to the question-raising mail if you would like to.)

The following are some questions I personally want to ask:

1. Memory and bandwidth problems can be simultaneously solved in the multiple master setting. However, the multiple master setting has been described in the CACM article, though, very briefly. So, I think what we do is to detail how masters work by using our idea?

2. Although namespace file partition the metadata in a static way, we may still directly modify the namespace file so that after a specific time, the metadata that belongs to master 2 before the specific time belongs master 3. So, maybe it is also dynamic?

Q1:
In the CACM article, they mention a "static namespace partition file" approach to enable multiple masters over a pool of chunkservers.
The description, however, is very rough and we are very interested in it.
So, we would like to describe that idea more clearly and identify its pro and con in the class.

Answer to Q1:

First, a static namespace partition file is just a mapping table of directory <-> master ID.
Clients will read this table to find the "right" master.
Master will write to this table for a directory creation request.

The corresponding master adds an entry to this table when a directory creation request arrives.
And we think the client chooses a master randomly to server this request.

The content of this table must be consistent to all clients and masters.
Therefore, we think the reasonable approach is only masters have this table.
And the scenario for a read request should be as follows.
1. Whenever a client send a read request, it chooses randomly a master and sends the request.
2. The master looks up the table for the corresponding master and redirect that request to it.
3. The corresponding master serves the read request and send the result back to client.
Therefore, each master has a "DISPATCH" functionality, i.e., to dispatch request to the right master according to the mapping table.

Things needed to be pointed out:
1. Although they call it namespace partition "file", we don't think this information is stored in file(in the disk).
It should resides in the memories of masters in order to keep faster response. After all, all requests must consult this table.

2. The granularity of partition is "directory". There are pro and con here.
pro 1: Large granularity results in less synchronization. Synchronization overhead can be small.
pro 2: This granularity simplifies the modification of the design of master.

con 1: Large granularity may result in load imbalance both in terms of memory usage and served requests.
Once a directory is created, all metadata of files and chunks in that directory will store in one master.
And all requests for all metadata of files and chunks in that directory will direct to one master.

3. Since they allow multiple masters over a pool of chunkservers. Then each chunkserver must maintain another table to map chunk ID to master ID.
So that the chunkserver knows which chunk should report to which master according to that table.

Concluding remarks:
The "static namespace partition file" approach for multiple masters is just a workaround approach for GFS.
The two pros makes multiple masters design to be done quickly and release the burden of single master design.
However, the con described above may be a problem.
So, what about more fine granularity: "FILE".
We will address this problem in our project proposal.




Q2:
Why the namespace partition file is "STATIC" rather than "DYNAMIC"?
Again, we would like to describe what we have discussed in the class.

Answer to Q2:

First, we give our definition about STATIC and DYNAMIC:
"STATIC" means once the entry (directory to master ID) is added into the file, this mapping won't change in the future.
On the other hand, "DYNAMIC" means a directory, say "/A", can be mapped to master 1 at a time, and then mapped to master 2.

The reason why not allow dynamic:
If we allowed the mapping to be changed, we must support metadata migration, which means to move metadata from one master to another master.
Since the granularity of partition is directory, one can imagine that there will be LOTS of metadata need to migrate when the mapping is changed.
So, the overhead of migration should be large since all requests to that directory must be suspend until the migration is done, and the mapping information of chunkservers should be change accordingly.
we believe the reason why they don't allow dynamic is that they want to simplify design since this is just a workaround approach.

Master Problem in GFS:
1' memory usage problem
2' bandwidth bottleneck
3' single failure problem

suppose we use two Metadata Server(MS)
we would like to have:
1' load balance: the metadata is evenly distributed among MSes

approach:
use dispatcher: each query request is sent to dispatcher who will decide which MS
is going to serve this request.

Then we have single dispatcher problems.
So our approach is to embed dispatch function into MS.

A user wants to create a file, then it sends creation request to arbitrary MS.
The MS first decide which MS is going to serve this request
with one mysterious algorithm. It decides MS1 should serve this request.
So it redirect the creation request to MS1.

As long as we can distribute creation request evenly, we can end balance

我需要：
一、 所有 speccpu各app所用到的instruction

一般的 instruction 的 translation function 不要處理 condition code，
可能嗎？這要再想想。

我還需要麼，
我需要的是：
1、一個 test program。

2、列出它用到的 opcode 及 operands 的值

3、qemu 停止translate 的 opcode 以及 它的處理方式。

4、目前code gen的 overhead是否高？

5、如何確定translation function寫對？

exec.c:910
tb_link_phy_page??????

今天的進度：
1、要完成 test case 的 qemu result
2、要檢驗目前完成的 qemu

call 的

正確性！
要如何確定正確性？
我本來是想要用 qemu 的 output 來當做正確性，
就是將 translate 過後的 inst 來比較
不過這方法只要會碰到 memory 就一定會 segmentation fault，
應該沒關係吧？

另外是有關 log file 的問題，exec.c: logfilename, 及 main.c 中有 DEFAULT_LOG_FILE
需要可以讓我指定檔名。

log file 何時開啟的？
加了 logfile 的 option，可是一定要在 -d 的後面，一定！

qemu stop conditions 有 58 個 instructions:

0059 addl $1000, %eax
0065 addl $1, %eax
0066 addl %gs:0, %edx
0272 call *%gs:16
0273 call *%eax
0277 call -100149
0467 cmpsb
0768 imull -4740(%ebp)
0809 int $128
0912 jae 1
0913 jae -135
0914 ja 1
0915 ja -1163
0916 jbe 100
0917 jbe -1281
0918 jb 10
0919 jb -156
0920 jcxz 44
0921 je 1
0922 je -10119
0923 jge 107
0924 jge -1081
0925 jg 100
0926 jg -1047
0927 jle 10
0928 jle -1177
0929 jl 10
0930 jl -1285
0931 jmpl *100(%ebx)
0932 jmpl *%eax
0936 jmp 10
0937 jmp -1007
0938 jne 10
0939 jne -1012
0943 jnp -213
0944 jns 10
0945 jns -1546
0948 jp 13
0949 jp -196
0950 js 10
0951 js -1081
0987 leal (%eax,%eax), %ecx
1274 movl %eax, %gs:(%edx)
1280 movl %gs:(%eax), %eax
1281 movl %eax, %ebx
1371 movsb
1372 movsl
1424 movzwl (%eax), %eax
1469 fmuls (%eax,%ebx)
1982 ret
1983 ret $12
2059 sarl %eax
2158 shll $4, -116(%ebp)
2197 shrl $11, %edx
2264 stosb
2265 stosl
2302 subl (%eax), %ecx
2303 subl %eax, %ebx

condition code is a challenge

我現在遇到的問題在於 debug，
我需要知道我是那一步錯了。
將 Makefile.target中的
#LDFLAGS+=-Wl,-shared
改成
LDFLAGS+=-Wl,-T,$(SRC_PATH)/$(ARCH).ld
就可以避免產生 PIE 的檔案。這樣就可以用 gdb 了。

現在遇到的另一個問題是：
要跳回qemu時的位置錯了！
就是我從qemu拿到的是 0x62933f88
可是經由 llvm 產生 call instruction 時的位置卻是0xdb24137。
而且這個值是會變的！

something wrong with "inttoptr" instruction

這是怎麼回事？

0x764, 1
28(Reg)

0x501, 2
28(Reg) 45(Reg)

0x4ff, 2
38(Reg) 5(Imm)

0x4ff, 2
27(Reg) 1(Imm)

0x501, 2
29(Reg) 38(Reg)

0x329, 1
128(Imm)

====================
現在好玩了，我發現 int 的處理是去 call helper_raise_interrupt
而非跳回去。
所以，只要 call 它就好了嗎？

call 了，可是跳不過去…

Question : how to link

可以了，哦耶！
要記得的教訓是：
1、要注意qemu對結束條件的處理，
像 int 0x80，qemu的處理法是呼叫 helper_raise_interrupt( 80, 02 )。
這要用 gdb 看才知道。
2、generated code的處理。LLVM的 CodeEmitter 將code產生在它自己管理的地方，
這部分要再處理一下。如，能否指定 memory 的位置，

目前已經可以產生 switch statement 以及 llvm translation functions header

我還需要麼，
我需要的是：
1、一個 test program。
2、列出它用到的 opcode 及 operands 的值
3、qemu 停止translate 的 opcode 以及 它的處理方式。
4、目前code gen的 overhead是否高？
5、如何確定translation function寫對？

exec.c:910
tb_link_phy_page??????

3/8-3/12

=======3/8========

現在的問題是如何解讀它 opcode？

第二個問題是為什麼compile出來的 qemu-i386是個 PIE？為什麼 gdb 會覺得它是？

我現在想幹嘛？

？？？？？？？？

instruction ID 到底是什麼？

=================

push %ebp
0x764, 1
28(Reg)

mov %esp,%ebp
0x501, 2
28(Reg) 45(Reg)

sub $0x8,%esp
0x8fd, 3
45(Reg) 45(Reg) 8(Imm)

movl $0x5,-0x4(%ebp)
0x4f9, 6
28(Reg) 1(Imm) 0(Reg) -4(Imm) 0(Reg) 5(Imm)

mov -0x4(%ebp),%eax
0x500, 6
27(Reg) 28(Reg) 1(Imm) 0(Reg) -4(Imm) 0(Reg)

mov %eax,(%esp)
0x4fa, 6
45(Reg) 1(Imm) 0(Reg) 0(Imm) 0(Reg) 27(Reg)

call 8048080
0x115, 1
-56(Imm)

===========================notes==========
1' the generated function contains a push in the first place, remove it!!!
2'

=======================================
use size_t to represent memory address size type

32bit VS 64bit, this is really a problem:w
:wG

ToDo

TODO:

1. take the course.

2. read papers

3. PP2010 judge system, worksheet items

4. Money:孫(10000)、王(5400)、包(5400)、房(6500)、信(9000)。

Mongrel Cluster Setting Steps

http://azureusonrails.rubyforge.org/wiki/wiki.pl?Install/Mongrel_Cluster_With_Apache_2.2

3/3

今天要看 paper，看那一篇好五了 -

Design and Engineering of a Dynamic Binary Optimizer.

看 openmp 的 worksheet，然後想一下如何用批改系統來改。

批改系統弄好了，

在 grid01.csie.ntu.edu.tw/~pp2010/judgegirl

3/1 - 3/5

過後的第二個星期，加油！喝！

加油，你可以的

另外，qemu 在 interrupt 之後就沒有再 disas code，why?

因為 syscall 的為 TARGET_NR_exit，所以就直接 _exit 了，

後面的 code 就不用 return 了。

首先：需要找出到底有那些 op 是我的 target？

先決定那些頁面要印，

我需要 instruction format 的 頁面，還有

REX prefixes are instruction-prefix bytes used in 64-bit mode. They do the following:

• Specify GPRs and SSE registers.

• Specify 64-bit operand size.

• Specify extended control registers.

要處理的 instructions 有：

83 c4 08 add $0x8,%esp

83 is op code, for ADD

c4 is ModR/M, 11 000 100 (2+3+3)

MOD: 11

Reg: 000

R/M:100

0000 1000

83 ec 08 sub $0x8,%esp

e8 c8 ff ff ff call 8048080

cd 80 int $0x80

5d pop %ebp

55 push %ebp

c3 ret

89 04 24 mov %eax,(%esp)

89 45 f8 mov %eax,-0x8(%ebp)

89 45 fc mov %eax,-0x4(%ebp)

89 cb mov %ecx,%ebx

89 e5 mov %esp,%ebp

8b 45 fc mov -0x4(%ebp),%eax

8b 4d fc mov -0x4(%ebp),%ecx

b8 01 00 00 00 mov $0x1,%eax

c7 45 fc 05 00 00 00 movl $0x5,-0x4(%ebp)

============3/2===============

每一個都要寫一個 disassembler

為什麼會這麼焦慮呢？

冷靜下來，好好的想，念「南無阿彌佗佛」，南無阿彌佗佛 南無阿彌佗佛 南無阿彌佗佛，肚子好餓，
translator 寫在一起算了！

重寫 cpu_gen_code!!

tb->tb_jmp_offset : 用來跳到 code cache 的其他地方，

./tcg/i386/tcg-target.c:873

tb->tb_next_offset : 用來跳到 code cache 的其他地方，

./tcg/i386/tcg-target.c:881:

s->tb_next_offset[args[0]] = s->code_ptr - s->code_buf;

這兩個跟我們應該沒關係，我們不會用到。

幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹幹

剛剛看了 tb_next_offset 的用途，並非沒用，

uint16_t tb_next_offset[2]; /* offset of original jump target */

uint16_t tb_jmp_offset[4]; /* offset of jump instruction */

unsigned long tb_next[2]; /* address of jump generated code */

tb_jmp_offset 在 ./exec-all.h:233 會用到，

offset = tb->tb_jmp_offset[n]，

這是在 tb_set_jmp_target 中，

tb_set_jmp_target(TranslationBlock *tb, int n, unsigned long addr)

先取出 tb 中 第 n 個 tb_jmp_offset，在該處放置於 addr 相對的位置：

jmp_addr = tb->tc_ptr + offset;

*(uint32_t*)jmp_addr = add - ( T + 4);

而 tb_set_jmp_target是在 tb_add_jmp中呼叫的的，

tb_add_jump(TranslationBlock *tb, int n, TranslationBlock *tb_next)

它先 check tb->jmp_next[n]不為0，若是，則呼叫 tb_set_jmp_target

然後將 tb->jmp_next[n] = tb_next->jmp_first;

然後將 tb_next->jmp_first = (TranslationBlock *)((long)(tb) | (n));

上面有個我不知道的假設，它似乎拿最後面兩個 bit 來用，

可是它怎麼知道 TranslationBlock *tb 指到的位置都沒有用到最後面兩個 bit，

那你要去看那個 pointer 是怎麼來的，tb_alloc，

沒什麼特別，是不是所有的 pointer都是最後兩個bit沒用？

是的，是的，pointer 的最後兩個 bit 都是沒用的，因為

一個 pointer 是4個byte(在 32 位元的機器上)。

在 cpu_exec 中會呼叫 tb_add_jump，

tb_add_jump((TranslationBlock *)(next_tb & ~3), next_tb & 3, tb);

上面function的意思是指：

若 next_tb->jmp_next[n]不為0，n is "next_tb & 3"

則將 tb->tc_ptr 跟 (next_tb->tc_ptr + next_tb->tb_jmp_offset[n])的距離，

放在 next_tb->tc_ptr + next_tb->tb_jmp_offset[n] 所指的位置上。

真的會用到嗎？

會！做實驗了。

從source code document中提到關於 jmp_next:

/* list of TBs jumping to this one. This is a circular list using

the two least significant bits of the pointers to tell what is

the next pointer: 0 = jmp_next[0], 1 = jmp_next[1], 2 =

jmp_first */

what does epilogue do?

restore stack and pop out saved register

回傳值是藏那去？

算了，先不要回傳值好了，

from wikipedia:

http://en.wikipedia.org/wiki/X86_calling_conventions

the returned value is stored in EAX register

來研究一下等會要幹嘛好了，

一、綀

=============3/3=================

一、prologue 跟 epilogue 都可用，

但要將 tb_ret_addr 要能 pass 過 translator。

另一方面，要想，有沒可能自己寫 prologue 跟 epilogue？

是要寫在 translation function 中嗎？不行吧，這樣你怎麼做 linking？

還要跳過 prologue 跟 epilogue。

test program 中有兩個 basic blocks:

55 push %ebp

89 e5 mov %esp,%ebp

83 ec 08 sub $0x8,%esp

c7 45 fc 05 00 00 00 movl $0x5,-0x4(%ebp)

8b 45 fc mov -0x4(%ebp),%eax

89 04 24 mov %eax,(%esp)

e8 c8 ff ff ff call 8048080

55 push %ebp

89 e5 mov %esp,%ebp

83 ec 08 sub $0x8,%esp

8b 45 08 mov 0x8(%ebp),%eax

89 45 fc mov %eax,-0x4(%ebp)

8b 4d fc mov -0x4(%ebp),%ecx

b8 01 00 00 00 mov $0x1,%eax

89 cb mov %ecx,%ebx

cd 80 int $0x80

要處理的 instructions 有

83 c4 08 add $0x8,%esp

83 is op code

08 is immediate value 0x8

c4 is ModR/M, 11+000+100 (2+3+3)

MOD: 11

Reg: 000

R/M:100

83 ec 08 sub $0x8,%esp

e8 c8 ff ff ff call 8048080

cd 80 int $0x80

5d pop %ebp

55 push %ebp

c3 ret

89 04 24 mov %eax,(%esp)

89 45 f8 mov %eax,-0x8(%ebp)

89 45 fc mov %eax,-0x4(%ebp)

89 cb mov %ecx,%ebx

89 e5 mov %esp,%ebp

8b 45 fc mov -0x4(%ebp),%eax

8b 4d fc mov -0x4(%ebp),%ecx

b8 01 00 00 00 mov $0x1,%eax

c7 45 fc 05 00 00 00 movl $0x5,-0x4(%ebp)

in translate-all.c, line 88

CPUX86State 很大，有 28536 byte，一個 page 才 4k=4096 byte，它就佔了7個 page！

==================================

先將 disassembler 弄好。

2/22~2/26

吃飯吧

接下來就要開始寫程式了

要如何才能專心在寫程式上呢？

你都要31歲了，難道不能自制一點嗎？

你不要用責備的方式，這樣是沒有用的，

如果罵一罵有話我現在搞不好已經畢業了呢！

不罵行嗎？講十天也沒用，就你不要看書了，還是硬要看完，

是怎樣？！那幹嘛專程來中研院看書，在家裡看不就好了，

what does "extern" mean in C++?

the following URL provides a good explanation:

http://developers.sun.com/solaris/articles/mixing.html

and, do I need to declare a extern "C" header file?

in llvm_translation.h, we should have a declaration as follows:

extern "C"{

typedef unsigned long (*PFunc_t)();

PFunc_t llvm_gen_code(CPUX86State *CPU1, const uint16_t *opc_buf, const uint32_t *opparam_buf);

}

now, goto qemu makefile

幹！專心啦

現在 make 是一個大問題

其實也還好，只要將 header include 進去，再 compile，再 link 就可以了，

llvm_gen_code要在那裡？

應該要放在 gen_code_slow 那邊吧:

========2/23============

工作工作，人生的意義就在工作裡

加一個target，llvm_translator 的 target，真想吃點東西，

Makefile.target，加在裡面:

$@: target name,

$<: dependent name

@CMD: execute CMD

寫個信給老師，說我禮拜三不會去，

幹！怎麼不吃大便，整天一直講話，

在 configure 中加入 llvm translator 的 config

可以在 configure 的時候決定是否要用 llvm_translator，

--enable-llvm-trans

要不要加在 config-host.mak。不用。

config-host.mak, config-host.h

只要加在 config.mak及config.h中即可。

config.mak, config.h

焦慮中

去洗臉吧

在LLVM中，JIT所產生的 code 是放在那？heap中嗎？還是mmap中呢？

有可能是在 heap 中，那會有什麼問題嗎？執行上的問題吧，

可是它可以執行啊；嗯，對，有可能是需要在 mmap 中，

ExecutionEngine::create 有另一個

create (

ModuleProvider *MP,

std::string *Err,

JITMemoryManager *JMM,

CodeGenOpt::Level OptLevel=CodeGenOpt::Default,

bool GVsWithCode=true)

其中 JITMemoryManager 就是控制 Memory 的 class，

但問題是，如何將把它跟 QEMU 的memory control弄在一起，

所以現在的問題是 qemu 是如何來控制 memory 的？

我猜是…，還真不知道，

我只知道滿了就清空這個，

乾脆將tb->tc_ptr = llvm_gen_code 算了，

這樣可能會死！死了再說！！！！！！！！

先能動！

走走走走走

gen_opc_buf: uint16_t buf[OPC_BUF_SIZE]

gen_opc_ptr: uint16_t *, move inside the buffer

又來了，幹，幹嘛一直看圖片啦，是在幹什麼！

下班下班，工作的意義在於下班！

===========2/24======================

今天要報告什麼呢？

上上禮拜五請假，禮拜四meeting，

報告我做的事好了：

1、將 整合進 qemu 的 Makefile 中，

2、指出 LLVM 是用 JITMemoryManager 來管理 translated code cache

但 qemu 是自己管理 translated code cache，在

qemu 的 code cache 管理很簡單，只是一直放 code，如果 buffer 滿了，就清空。

cpu_gen_code 需要return gen_code_size，

因為它需要 advance code_gen_ptr；code_gen_ptr指到目前可以放code的位置。

regs_to_env is empty in i386

translated code alignment problem in QEMU,

QEMU forces all generated code to be 16-bit aligned!!!!

IR Cache - 這會存在 Module 中，

Basic Block Cache，這會存在 JITMemoryManager 中，

所以如果以後要做 Code Cache Memory 的研究，應該 extend 這

=========2/25=============

今天要做什麼呢？

可以做 translate.c 裡的 disas_insn_llvm 這個 fuction

也可以寫一個 translation function，

可是你還沒訴 Makefile 可以找到 llvm-translator.cpp 這個檔，do it！

結果只是 rules 寫錯，幹！

加入 llvm 的檔案，compile 需要，

cscope -R -b

整理一下，

今天已成功將 llvm translator integrate 到 qemu 的 make system

遇到的困難有，c 與 c++ 檔案的 link 的問題，

主要是我想在 llvm-translation.h 中將 llvm_gen_code_wrap 這個程式用 static 的

方式用在 translate-all.c 中，

不過好像不應該寫在 header file 中，直接寫在 translate-all.c 中。

but，anyway，想講的是，如果在 extern "C" 中要用到 static，就直接寫在裡面就好了，

如：

extern "C" {

static void* f (void){ /* write your code here */}

}

不要這樣：

extern "C" {

static void* f (void);

}

extern "C" static inline void* f(void)

{

/*write yout code here*/

}

上面的方式會有 linkage error 的錯誤訊息。

另外，今天也將 llvm 加入 svn 系統中，這樣就將 llvm 的版本固定下來了。

在 xeon 上跟 haru 上都可以打 make 編譯了。

做別的事轉換一下心情吧。

看paper。

NEW YEAR TODO

2010 2.8 note

一、我是否要用 llvm project 來建我的東西？

嗯，我覺得不一定要用它的東西。可以考慮，不過不是現在要，先知道有這個東西。

LLVM實在是一個很好的project，很難相信他們可以寫這麼多的document。

docs/ProgrammersManual.html

包含很多使用 llvm 寫程式的實用知識，

今天要將 llvm-translation.h 編好，能不能用倒是其次，要能 compile 倒是真的。

去大便吧！

compile llvm ... done, installed in /home/tk/local/llvm-2.6

然後compile llvm-gcc ... done, installed in /home/tk/local/llvm-gcc-2.6

去洗個臉，然後找啟彰去走走。

去洗個臉。

照理說，dyngen 應該是要用程式寫出來的，

合理的想法是，先寫 llvm-database，然後再parse database裡的東西寫出。

不過這我目前沒法做到，

老師也沒預期我會這樣做，所以還是先用手寫吧。

廢

即然不想做事，就來寫點東西好了，主要是今天看到的：

1、Big-Endian V.S. Little-Endian

ON HOLY WARS AND A PLEA FOR PEACE by Danny Cohen

Danny Cohen 是 ADOBE 的兩位創始者之一，PostScript是他們發明的。

這篇是寫電腦科學中有名的問題，此君以幽默的語氣完整的介紹了此問題，

十分有趣，值得一讀。

2、

TODO:

1' find out <- this is llvm translation database
2' find out the translator engine <- llvm-op.h is the engine
3' add translator engine into llvm_gen_code
4' in the slide, you should mention the architecture of the framework, and
indicate which is done.
5' show the target instructions and say that you are going to build the translation function in the next week

QEMU programming notes

profiler => can we ignore it?

inhibit irq => true if hardware interrupts must be disabled for the next instruction

TF cpu flag => 8th bit of flags in x86, trap flag, system flag

jmp_opt in DisasContext => indicate whether to do block linking

what is GRP1? in

what is modrm?

Q: what are my target instructions?

test program is in ~/research/qemu/src/branches/llvm-qemu/tests/

add-i386.c

tmp.c

stay in disas_insn() in target-i386/translate.c

line 4122

I need a C programming language book

1' create a new fucking useless function!

mov

0x89 source is register, destination could be register

or memory location related to a register

svn merge

svn merge -r 5600:5646 ../branches/march

running qemu as daemon

I am using another way:

qemu -hda /qemu/xp.cow -boot c -cdrom /data/install/admin/os/windows/winxpsp3ue.iso -m 384 -k es -vnc *:0 -daemonize -localtime

-vnc *:0 to run a vnc server wich I can connect from anywhere (that's the asterisk for - *) on the display 0

-k es when using VNC instead of SDL you have to specify the keyboard layout, I am using spanish

-daemonize this keeps the process running as a daemon.

-localtime with this clock is set to the hardware clock, not utc. This is for windows

So, when i want to connect back i just use

grep exclude pattern

grep -ircl --exclude=*.{png,jpg} "foo=" *

grep -Ir --exclude="*\.svn*" "pattern" *

TODO

TODO:

1. 打電話給吳小姐
2. 將訃文拿給人事
3. 注意實驗的進度
4. 做translation function
   

enable IBM trackpoint and disable touchpad

first, install xinput: sudo apt-get install xinput

xinput list => show all input devices
xinput list-props $dev_id => show all options of $dev_id
xinput set-int-prop $dev_id "$opt" 8 $value => set integer value to option $opt,

xinput set-int-prop $touchpad "Device Enabled" 8 0
xinput set-int-prop $trackpoint "Evdev Wheel Emulation" 8 1
xinput set-int-prop $trackpoint "Evdev Wheel Emulation Button" 8 2
xinput set-int-prop $trackpoint "Evdev Middle Button Emulation" 8 1

meeting結論

1、叫學弟做實驗分析QEMU的執行時間：translation, execution, control。
2、用 ARM 跑PARSEC2.0，可能要cross compile
3、將QEMU x86上不能跑超過三個threads回報至mail中。

ToDo

一、藝鴻
1、流程要能跑完。
二、unigrid demo
1、要跟晁睿拿 ibm
2、後製要到我的電腦上用 k9 來做
三、師大的案子(星期三)
1、電子報上，
2、討論區中譯，
3、手冊？
四、民航局的案子
1、阿碼科技的檢驗提案。
2、pixy的問題
五、JSS
1、Lovasz Local Lemma 的 implementation
2、JSS演算法的了解
六、Binary Translator
1、LLVM

一、寄通知信給各老師，請他們啟動 SRB
二、做 slide，demo sequence
三、

Transitive's QuickTransit

Apple Computer implemented a dynamic translating emulator for M68K code in their PowerPC line of Macintoshes, which achieved a very high level of reliability, performance and compatibility (see Mac 68K emulator). This allowed Apple to bring the machines to market with only a partially native operating system, and end users could adopt the new, faster architecture without risking their investment in software. Partly because the emulator was so successful, many parts of the operating system remained emulated. A full transition to a PowerPC native operating system (OS) was not made until the release of Mac OS X (10.0) in 2001, and within this new OS the "Classic" runtime environment still offers the emulation capability on PowerPC Macs. Also, the Rosetta translation layer included in releases of Mac OS 10.4 for Intel-based Macs, which is used to ease the transition from the PPC to x86, is an example of dynamic translation. Developed for Apple by Transitive, the Rosetta software is an implementation of Transitive's QuickTransit solution, which can be used to dynamically translate between platforms that include SPARC, PowerPC, MIPS, Itanium and x86.

Open64 Research - Work item

1. read whirl node

Open64 - cg

Questions:

在那 fprintf assembly

ToDo

一、問阿斌封面的事
二、看一下 無障礙 tab 鍵
三、OPEN64
四、師大

TODO

一、啟彰的 slide
二、明早 meeting 的 slide
三、明天下午的 slide
四、後天 meeting 的 slide
五、annotation的東西
六、明天早上打給黃雲英小姐
七、明天記得帶 dvd 出門

ToDo

1、明天要去航管局資管中心蓋章
2、明天早上要來中研院開會
3、要還 dvd，出門記得帶著
4、確定六項的名稱，條碼，明天要生海開始檢查，要告訴他如何檢查
5、要記得問真慧版，跟師大聯絡
6、跟藝鴻討論一下，順便拿給簽好的東西
7、記得買保險套
8、要記得跟德禹講錢的事
9、明天要聚餐，台北 101 四樓 Wasabi
10、

php mailer gmail

要先跳過去，
要有menu，

$phpmailer->CharSet = 'utf-8';
$phpmailer->Encoding = 'base64';


//$phpmailer->Sender = "cher.webmaster@gmail.com";
//$phpmailer->Sender = "tk@sage.csie.ntu.edu.tw";
//$phpmailer->IsSMTP(); // set mailer to use php mail()
//$phpmailer->Host = 'ssl://smtp.gmail.com';
//$phpmailer->Port = 465; //default is 25, gmail is 465 or 587
//$phpmailer->SMTPAuth = true;
//$phpmailer->Username = "tp.killer@gmail.com";
//$phpmailer->Password = "ix jn tk";

OPEN64 - Glossary

TY = type
ST = symbol table
TN = temporary node
PREG = pseudo register

Question about ianthro

Question:

I still don't know how to find out where 'navigation' is generated.

drupal path fix

1. /collections has been URL-aliased to node/2779. Therefore, we won't need
"collections " in menu system(remove $items['collections'])
1.1 remove "$items['collections']" out of menu system will lost 'collections' in navigation bar, don't know why

2. we don't need function collections_nodeapi since now /collections has been mapped to node/2779. We don't need it to render '/collections' page

3. remove all loader since now we use page callbacks to connect collections and drupal

4. the function of collection_preprocess_page. It will append $collections['template_files'] to $variables['template_files'] which will be used to generate collections pages in theme().

4.1. We need the condition if ( ! isset($collections) ) since collection_preprocess_page could be called without being followed _collections_load which we define collections.

工作日誌：

ianthro：安裝 phpMyAdmin 及 lynx
gogogogogogogogogogogogogogogo

中研院管考

正式系統：
tag/march/0.01
1. chart disabled; comment out chart button in SelectPanel.mxml
2. svn commit -m ""; svn cp CURRENT
3. svn co <>
3. add config/database.yml
4. run march: ./start_march restart 140.109.136.136
5. mv db/migrate/032xxx .; rake db:migrate; mv 032 back
6 add operations operation_items uploaded_files data into database
7. cp IHP/bin/pdftk to production directory

TODO

一、寫作業。
二、統計 worksheet
三、交心得報告
四、藝鴻的東西, http://sb.ianthro.tw/
五、greedy algorithm，approximation
六、

hadoop cluster and eclipse plugin

Eclipse : 3.3.2
Hadoop: 0.19.1
Hadoop-eclipse-plugin: 0.19.1, in contrib/eclipse-plugin

cluster environment:
grid01(cluster head) has two network interfaces:
140.112.30.240 --> external
192.168.77.1 --> internal

grid02~grid05 are in the private network:
192.168.77.2~5.

Since we need to contact the internal nodes when running hadoop applications,
we need to setup proxy to have grid02~grid05 accessible to people outside.

Setup proxy port 54088:

ssh -L 140.112.30.240:12345:192.168.77.1:22 140.112.30.240 -n -N &
ssh -D 140.112.30.240:54088 -p 12345 140.112.30.240 -n -N &

And then in hadoop-eclipse plugin, you can set proxy host/port as 140.112.30.240:54088

memo

1, cvs check, user upload
2, month

work

Work!!!!

funwish todo

一、月份不是從 1 月開始
二、csv 上傳時需要自動 insert，actual，expect，digital_output，
三、測試報告

memo

need to judge when to put final
but right now, only need to consider season

gogogo

second round, GO

what about digital?

16 2 3 7 8 10 1

TODO

要做的事有：
一、申請簽證，明天早上九點半
二、訂 Frankfurt YH，
三、準備 Hadoop worksheet
四、setup hadoop machines
五、research，pin，dynamoRIO，XED2，qemu。

build UQBT

build uqbt

require libelfg0-dev package

1. tar xzf uqbt1.0alpha.tar.gz
2. ./configure --with-source=pent --with-target=sparc --disable-vpo --enable-remote --prefix=/home/tk/local
3. vim Makefile
change CPPFLAGS= -I/usr/bin/../include/linux to CPPFLAGS=
change CC=gcc to CC=gcc-2.95
change CXX=g++-2.95
pass '--traditional-format' option to linker
%s/-lstdc++/-Wl,--traditional-format -lstdc++/g
4. vim include/config.h
change #define HAVE_ELF_H 1 to #define HAVE_ELF_H 0
5. vim include/ElfBinaryFile.h
put #include to the top of #ifndef ELF32_R_SYM
6. vim loader/ElfBinaryFile.cc
elf_hash(pName) ==> elf_hash((unsigned char*)pName)
7. make

The processes goes fine until ld, there will be alot of errors shown, put we still can get the uqbtps executable file

however, it cannot translate the compiled program ...

我是不是有病

我是不是有病

要很認真的思考這個問題，我是不是有病！

無法專心，無法完成一件事，
整個頭腦亂成一團，我需要整理我的腦袋。
定期的

Setup OpenSuse

1' setup network: /etc/sysconfig/ifcfg-eth0 /etc/sysconfig/ifroute-eth0
2' setup firewall, sudo /sbin/yast, choose firewall, allow SSH, and related services to through firewall
3' set NIS server
4' set NFS server
5' set NIS client
6' set NCF client

TODO

1' Announce TA office hour
2' study OpenMP, CUDA
3' Prepare power extending lines for classes
4'

TODO

1' understand OPEN64 CGIR form
2' collect binary transformation, binary optimization papers
3'

delete in operation_item_controller.rb has problem, when updating items' ranks

model cannot use type as attribute name

Cannot use "type" as attributed name in model.

it will issue an error when you use find, e.g., Operation.find :all
>> Operation.find 1
SyntaxError: compile error
/home/tk/eclipse/flex/ihp-report/branches/march/IHP/vendor/rails/activerecord/lib/active_record/base.rb:1748: syntax error, unexpected tINTEGER
Object::0

shit

今天要做的事

一、看曉偉的東西，
二、experiment about that "does greedy algorithm produce optimal solutions for file transfer problem."
三、artit 的東西。
四、funwish 的東西。
五、multi-core meeting, tutorial
六、

2nd open 64 tutorial

where does mcf-seq-f and mcf-par-f come from? are they copied from the outputs of 1st tutorial ?

鏡子

鏡子，今天要去買鏡子。
昨天跟依親聊天時她的匿稱也是鏡子，剛看msn的列表時也有人是到鏡子。
這是什麼預兆呢？

xrandr

interesting stuff : xrandrx

xrandr --auto ( auto-detect newly add screen(VGA, another monitor, projctor, etc.))

xrandr --output VGA-0 --right-of LVDS ( create a big screen where VGA-0 is right of LVDS )

an optimal solution for edge-contention is an optimal solution for out-degree? or vice versa?

There is a simple question:

an optimal solution for out-degree-minimization is an optimal solution for edge contention?

If we say a solution is optimal for edge-contention, is it an optimal solution for out-degree-minimization?

要做的

寫程式測試 file transfer problem 的 greedy 的演算法是否有反例

試 open64

REVIEW 曉偉的 paper

要閱讀的

Using Open64 based TLS compiler

Pin: Building Customized Program Analysis Tools with Dynamic Instrumentation
SuperPin: Parallelizing Dynamic Instrumentation for Real-Time Performance

Hill-Conf. Presentation Advice.pdf
Hill-Conference Etiquette.pdf
Hill-Writting.Tips.pdf
How to be a winner.pdf