[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 28 Jul 2013 20:28:32 -0400
From: Yaniv Sapir <yaniv@...pteva.com>
To: john-dev@...ts.openwall.com
Subject: Re: Parallella: bcrypt
Katja,
Just a piece of debugging advice - you can embed or insert an "idle"
instruction anywhere in the code.Once the core is idle, it is then save to
probe the contents of the GP registers (R0-R63). Use
> e-read 0 0 0xf0000 63
to list all the GP regs and inspect their contents at the idling point (of
course, you can list just the required regs instead). You should then be
able to spot a discrepancy in a specific register, whose purpose you
recognize.
Other than that, I'll second Alexander's advice - I believe that at this
stage you will have better control over the code if you program move to
assembly. To have an initial point, you can use e-gcc's assembly emitted
code, generated form the working C code:
e-gcc -S mycode.c -o mycode.s
then change your build line:
e-gcc -T ... mycode.c mycode.s -o mycode.elf
Yaniv.
On Sun, Jul 28, 2013 at 8:04 PM, Katja Malvoni <kmalvoni@...il.com> wrote:
> Hi Yaniv, Alexander,
>
> I have following code which is supposed to replace do while loop in
> BF_encrypt2():
>
> unsigned int offs1, offs2;
> offs1 = ctx1->s.P - ctx0->s.P;
> offs2 = (ctx1->s.P - ctx0->s.P) + 1;
> __asm__ __volatile__(
> "loop2: add r45, %[ctx0], 0x4\n"
> "add r46, %[ctx1], 0x4\n"
> "eor %[L0], %[P00], %[L0]\n"
> "eor %[L1], %[P10], %[L1]\n"
> BF2_2ROUND
> BF2_2ROUND
> BF2_2ROUND
> BF2_2ROUND
> BF2_2ROUND
> BF2_2ROUND
> BF2_2ROUND
> BF2_2ROUND
> "eor r22, %[R0], %[P017]\n"
> "str r22, [%[ptr]]\n"
> "str %[L0], [%[ptr], +0x1]\n"
> "eor r23, %[R1], %[P117]\n"
> "str r23, [%[ptr], +%[offs1]]\n"
> "str %[L1], [%[ptr], +%[offs2]]\n"
> "add %[ptr], %[ptr], 8\n"
> "sub r24, %[end], %[ptr]\n"
> "mov %[R0], %[L0]\n"
> "mov %[L0], r22\n"
> "mov %[R1], %[L1]\n"
> "mov %[L1], r23\n"
> "bgtu loop2\n"
> );
>
> The problem is that I get more or less same objdump (differences are in
> assigned registers and irrelevant differences in instructions order) with
> loop being in C and loop being in assembly but when loop is in assembly I
> get incorrect result. I spent a lot of time trying to figure out where is
> mistake and I can't find it. But I need a way to preload P arrays in
> assembly and implementing whole loop in assembly was my only idea. The same
> approach for implementing loop works in BF_encrypt(). If I copy paste code
> from that loop, I get incorrect result too.
>
> I pasted below only relevant lines of both objdumps i.e. last update of
> used register (except for L0, L1, R0, R1 which are updated in BF2_2ROUND)
> and parts before and after BF2_2ROUND. I can't find difference in these two
> objdumps that would cause one producing correct results and another one
> incorrect.
>
> When loop and storing Ls and Rs is in C compiler generated code is:
> sub r53,r1,r0 ; sub r53, ctx1, ctx0
> asr r53,r53,0x2
> lsl r57,r53,0x2 ; r57 = offs1
> add r58,r57,4; r58 = offs2
> [...]
> add r3,r0,72
> mov r2, r3 ; r2 = ptr
> [...]
> ldr r4,[r0,+0x11] ; r4 = P0[17]
> ldr r59,[r1,+0x11] ; r59 = P1[17]
> [...]
> mov r49,0x1048
> add r49,r0,r49 ; r49 = end
> [...]
> lsr r22,r48,0x6
> eor r12,r12,r23
> lsr r62,r47,0x6
> eor r26,r26,r63
> [...]
> 1818: mov r48,r12 ; mov R0, L0
> mov r47,r26 ; mov R1, L1
> mov r12,r22 ; mov L0, r22
> mov r26,r23 ; mov L1, r23
> add r45,r0,4
> add r46,r1,4
> [...]
> eor r22,r48,r4 ; eor r22, R0, P0[17]
> str r22,[r2] ; str r22, ptr
> str r12,[r2,+0x1] ; str L0, ptr+1
> eor r23,r47,r59 ; eor r23, R1, P1[17]
> str r23,[r2,+r57] ; str r23, ptr+offs1
> str r26,[r2,+r58] ; str L1, ptr+offs2
> add r2,r2,8 ; add ptr, ptr, 8
> sub r24,r49,r2 ; sub r24, end, ptr
> bgtu 1818 <__HALF_BANK_SIZE_+0x818>
>
> When loop is in assembly:
>
> sub r54,r1,r0 ; r54 = offs1
> add r57,r54,4 ; r57 = offs2
> [...]
> add r18,r0,72 ; r18 = ptr
> [...]
> ldr r55,[r0,+0x11] ; r55 = P0[17]
> ldr r56,[r1,+0x11] ; r56 = P1[17]
> [...]
> mov r17,0x1048
> add r17,r0,r17 ; r17 = end
> [...]
> lsr r22,r3,0x6
> eor r12,r12,r23
> lsr r62,r2,0x6
> eor r16,r16,r63
> [...]
> loop2: add r45,r0,4
> add r46,r1,4
> [...]
> eor r22,r3,r55 ; eor r22, R0, P0[17]
> str r22,[r18] ; str r22, ptr
> str r12,[r18,+0x1] ; str L0, ptr+1
> eor r23,r2,r56 : eor r23, R1, P1[17]
> str r23,[r18,+r54] ; str r23, ptr+offs1
> str r16,[r18,+r57] ; str L1, ptr+offs2
> add r18,r18,8 ; add ptr, ptr, 8
> sub r24,r17,r18 ; sub r24, end, ptr
> mov r3,r12 ; mov R0, L0
> mov r12,r22 ; mov L0, r22
> mov r2,r16 ; mov R1, L1
> mov r16,r23 ; mov L1, r23
> bgtu loop2
>
> Kajta
>
--
===========================================================
Yaniv Sapir
Adapteva Inc.
1666 Massachusetts Ave, Suite 14
Lexington, MA 02420
Phone: (781)-328-0513 (x104)
Email: yaniv@...pteva.com
Web: www.adapteva.com
============================================================
CONFIDENTIALITY NOTICE: This e-mail may contain information
that is confidential and proprietary to Adapteva, and Adapteva hereby
designates the information in this e-mail as confidential. The information
is
intended only for the use of the individual or entity named above. If you
are
not the intended recipient, you are hereby notified that any disclosure,
copying,
distribution or use of any of the information contained in this
transmission is
strictly prohibited and that you should immediately destroy this e-mail and
its
contents and notify Adapteva.
==============================================================
Content of type "text/html" skipped
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
