API tools faq
paste
Vulpes

float_asm_discussion

Vulpes
Nov 30th, 2011
145
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 12.94 KB | None | 0 0
raw download clone embed print report
  1. [12:27] • Vulpes raises hand:
  2. [12:27] Vulpes: if I empty off my floating register-stack, will it affect other variables that I have set outside my assembly code?
  3. [12:29] Vulpes: ie, if I: main(){ double a=1.0f; assembly_call_erases_fpstack(); printf("%f \n",a); etc...
  4. [12:29] Vulpes: in my code, a is now NAN. was the float value never saved to memory?
  5. [12:30] Jester01: maybe. check the assembly code. and the calling convention documentation :)
  6. [12:33] Jester01: generally you should only pop values you pushed
  7. [12:34] iiyo ([email protected]) joined the chat room.
  8. [12:36] Jester01: for the record, my gcc moves the variable initialization to after the assembly call
  9. [12:37] • Jester01 tries passing "a" as argument
  10. [12:37] Vulpes: I can link you my code, but I'm not sure you want to get your hands this dirty
  11. [12:38] Jester01: sure :)
  12. [12:39] arek_deepinit ([email protected]) joined the chat room.
  13. [12:39] • Vulpes looks at his .c
  14. [12:39] zach (817442c2@gateway/web/freenode/ip.129.116.66.194) joined the chat room.
  15. [12:39] Vulpes: wait ... hold up
  16. [12:39] AlbertoSilva ([email protected]) left the chat room. (Read error: Connection reset by peer)
  17. [12:39] Jester01: no matter what I did so far, gcc always called other functions with empty fpu stack
  18. [12:40] AlbertoSilva ([email protected]) joined the chat room.
  19. [12:42] monsterwizard ([email protected]) left the chat room. (Read error: Connection reset by peer)
  20. [12:42] FUZxxl ([email protected]) left the chat room. (Quit: euler=:[:+/i.*[:+./0=3 5|"0 1 i.)
  21. [12:42] zach: i have something strange going on when I assemble something with nasm. The line verbatim is "BootDrive equ bp-2" but it gives the error "invalid operand type." Help?
  22. [12:44] Jester01: that's invalid because EQU is trying to define a symbol with a known value. your value is dependent on value of the BP register at runtime. if you want simple text replacement, use a macro
  23. [12:44] Jester01: such as %define BootDrive bp-2
  24. [12:47] zach: ah, makes sense, thanks
  25. [12:47] Philopon ([email protected]) joined the chat room.
  26. [12:48] Vulpes: links:
  27. [12:48] Vulpes: http://pastebin.com/5q00PFs6 http://pastebin.com/BC8tmwx1
  28. [12:48] Vulpes: not necessary to view: http://pastebin.com/5NGJHAvJ http://pastebin.com/wc9JvVLr
  29. [12:49] Vulpes: so all I'm doing is loading all the floats up into the stack so I can print them out, which works
  30. [12:49] Vulpes: then I attempt to load them all back in
  31. [12:49] wobster ([email protected]) left the chat room. (Quit: wobster)
  32. [12:49] Vulpes: then I'm done ... or so I was to think so
  33. [12:51] Vulpes: on that 4th link, lines 144-151 print out the floats on the stack
  34. [12:51] Jester01: your asm code is wrong, but should not affect the C code
  35. [12:51] Vulpes: which were put in there on 2nd link lines 36-45
  36. [12:52] Vulpes: which places?
  37. [12:55] Vulpes: 37 & 39 are supposed to be dwords, right? not qwords? I'd assume so because they print out correctly
  38. [12:55] repozitor ([email protected]) left the chat room.
  39. [12:56] Jester01: you are popping 8 values from the stack, even if there were not that many present
  40. [12:56] Jester01: then you push them back
  41. [12:56] Jester01: so you will always return with a full stack
  42. [12:57] Vulpes: is there something wrong with that?
  43. [12:57] Jester01: furthermore, you are adjusting the stack pointer every time, so what is line 173 doing?
  44. [12:57] Jester01: YES
  45. [12:57] Jester01: as I said, you are not restoring the fpu stack
  46. [12:58] Jester01: you always pop off 8 values, even if there were not that many
  47. [12:58] Vulpes: oh sorry 173 sould be commented out
  48. [12:58] Jester01: and you always push 8 back
  49. [12:58] Jester01: suppose stack only had 1 value
  50. [12:58] Jester01: you pop 8 times
  51. [12:58] Jester01: then push back 8 items
  52. [12:58] Jester01: so now stack has 8 items instead of 1
  53. [12:59] Jester01: your test code does fill it with 8 values
  54. [12:59] Jester01: so it will work in that case
  55. [12:59] Vulpes: is there some sort of counter index that shows how many items are on the stack?
  56. [12:59] Burlingk ([email protected]) left the chat room. (Read error: Connection reset by peer)
  57. [12:59] Vulpes: I thought if you pushed too much it would fall off or popped too much, you would get garbage values, but nothing else
  58. [13:00] Vulpes: I figured if I popped everything off the float stack & then pushed it all back on, regardless if I took garbage numbers, all the data would be the same
  59. [13:00] Vulpes: am I not understanding instructions correctly?
  60. [13:02] Reversy ([email protected]) left the chat room.
  61. [13:03] plusk ([email protected]) joined the chat room.
  62. [13:05] nANdy (3ec730fd@gateway/web/freenode/ip.62.199.48.253) joined the chat room.
  63. [13:06] Jester01: hm you may be right about that (except for the stack overflow detection)
  64. [13:06] Jester01: stack top is stored in the fpu status register, btw.
  65. [13:07] • Vulpes learned something new
  66. [13:08] Jester01: HOWEVER .. FPU stack may be irrelevant to the current problem .. since you are not saving general purpose registers that you should definitely preserve
  67. [13:09] killown (~geek@unaffiliated/killown) joined the chat room.
  68. [13:09] Jester01: notably, ESI
  69. [13:10] Vulpes: ah, is my problem that my entire asm function isn't callee saving?
  70. [13:10] Vulpes: bbrb trying out
  71. [13:12] Jester01: seems to work here
  72. [13:13] Vulpes: eh? could I trouble you for your code?
  73. [13:13] Jester01: I just added push/pop esi to your test_floats and removed the unnecessary esp adjust
  74. [13:13] Jester01: note that the value of "c" is not preserved
  75. [13:14] Vulpes: that is fine.
  76. [13:14] Jester01: yah
  77. [13:14] Jester01: oh, and I have renamed your custom .txt to .text ;)
  78. [13:14] Jester01: which is probably irrelevant
  79. [13:15] Vulpes: I'm fuzzy on those pseudo ops, (not sure which fork to eat salad with)
  80. [13:15] Jester01: http://pastebin.com/uhgWMX9C
  81. [13:18] Vulpes: hmm
  82. [13:18] Vulpes: this is an interesting scenario
  83. [13:19] Jester01: I have checked, gcc here DID expect ESI to be unchanged
  84. [13:20] Vulpes: I wonder if it is the fact that I am running a VM
  85. [13:21] Vulpes: http://i.imgur.com/k60e9.png
  86. [13:22] Jester01: what happened with the code I pasted?
  87. [13:22] Jester01: crash? bad number?
  88. [13:22] Vulpes: xBFC82F38: -NAN
  89. [13:22] Vulpes: xBFC82F3C: -NAN
  90. [13:22] Vulpes: same rewrites to variables a & b
  91. [13:23] Jester01: http://pastebin.com/yHDdRQH4
  92.  
  93. [13:24] Vulpes: thanks for that output paste, I am going to see if I can dissect that, brb
  94. [13:25] zach (817442c2@gateway/web/freenode/ip.129.116.66.194) left the chat room. (Quit: Page closed)
  95. [13:39] Jester01: or you can try this binary http://filesmelt.com/dl/a.out_.gz
  96. [13:39] Jester01: (static)
  97. [13:45] rendar ([email protected]) left the chat room.
  98. [13:46] Horgh^ ([email protected]) left the chat room. (Quit: ++)
  99. [13:49] Vulpes: thanks Jester01 your binary is perfect
  100. [13:50] Vulpes: I wonder if it's my arguments
  101. [13:50] Hack004 ([email protected]) left the chat room. (Ping timeout: 252 seconds)
  102. [13:50] Jester01: can you send me your binary?
  103. [13:50] Vulpes: ok hold up
  104. [13:50] Jester01: also, you can put a write watchpoint on the address of a or b in gdb
  105. [13:50] lane ([email protected]) joined the chat room.
  106. [13:50] Jester01: to catch who is overwriting it :)
  107. [13:53] Vulpes: http://filesmelt.com/dl/div_test.out_.tar.gz
  108. [13:55] Jester01: hmmm
  109. [13:55] Vulpes: you get the NANs?
  110. [13:55] Jester01: that only prints the registers for me
  111. [13:55] Jester01: no
  112. [13:55] Jester01: I only get a single general purpose register dump :-O
  113. [13:56] Vulpes: it doesn't even print the stack?
  114. [13:56] Vulpes: did it segfault?
  115. [13:56] Jester01: no
  116. [13:56] Jester01: and no
  117. [13:56] Vulpes: http://pastebin.com/sguEB2Tn
  118. [13:56] Jester01: it doesn't even print the *initial* C stuff about a,b,c
  119. [13:57] slidercrank ([email protected]) left the chat room. (Quit: NO CARRIER)
  120. [13:57] slidercrank ([email protected]) joined the chat room.
  121. [13:57] Jester01: http://pastebin.com/Zs67gJnQ
  122. [13:57] Jester01: that's all
  123. [13:58] Jester01: can you create a static version? just pass -static to gcc
  124. [13:58] Vulpes: 1 & 3 loaded in ecx & edx ... where they were supposed to have been loaded with ebp & esp before the subroutine call
  125. [13:58] Vulpes: I don't even.
  126. [13:59] Griwes (~griwes@unaffiliated/griwes) left the chat room. (Ping timeout: 244 seconds)
  127. [14:01] Philopon ([email protected]) left the chat room. (Quit: Philopon)
  128. [14:03] plusk ([email protected]) left the chat room. (Quit: Leaving)
  129. [14:07] lane ([email protected]) left the chat room. (Ping timeout: 252 seconds)
  130. [14:08] arek_deepinit ([email protected]) left the chat room. (Read error: Connection reset by peer)
  131. [14:09] Jester01: hm, valgrind has trouble with "enter"
  132. [14:09] • Jester01 gets rid of it ... the enter, not valgrind :)
  133. [14:10] LunaVorax ([email protected]) joined the chat room.
  134. [14:10] Vulpes: hmm, maybe I really should forsake "enter" & go with the push ebp ; mov ebp esp
  135. [14:11] Jester01: ==22766== Use of uninitialised value of size 4
  136. [14:11] Jester01: ==22766== Conditional jump or move depends on uninitialised value(s)
  137. [14:11] Jester01: both inside printf
  138. [14:11] Vulpes: oh you know what
  139. [14:12] Vulpes: I disregard scope when printing out the stack
  140. [14:12] Jester01: ah yes
  141. [14:12] Jester01: that looks harmless
  142. [14:12] Jester01: and no other errors reported
  143. [14:12] Reversy ([email protected]) joined the chat room.
  144. [14:12] Smartnow (~Smartnow@unaffiliated/smartnow) joined the chat room.
  145. [14:12] Reversy ([email protected]) left the chat room.
  146. [14:12] Jester01: so, do you create a static version for me?
  147. [14:13] Jester01: maybe libc differences make your version not run here
  148. [14:13] Jester01: but first get rid of the "enter" instructions :)
  149. [14:14] SunnyD ([email protected]) joined the chat room.
  150. [14:15] Vulpes: I'm a bit unsavvy with the term. is that an assembly argument?
  151. [14:15] Vulpes: assembler*
  152. [14:15] Vulpes: also I made up that "unsavvy" word & realized it is not in the dictionary
  153. [14:19] Vulpes: I presume: gcc -static-libgcc
  154. [14:21] ImTheBitch ([email protected]) joined the chat room.
  155. [14:22] Jester01: no, just -static
  156. [14:22] jadugar (~jadugar@gateway/tor-sasl/jadugar) left the chat room. (Quit: leaving)
  157. [14:24] Jokke (6d5af7d9@gateway/web/freenode/ip.109.90.247.217) left the chat room. (Ping timeout: 265 seconds)
  158. [14:25] Vulpes: http://filesmelt.com/dl/div_test.out_.tar1.gz
  159. [14:25] Jester01: ahhhhha! got the nans :)
  160. [14:26] BoarderX ([email protected]) left the chat room. (Ping timeout: 260 seconds)
  161. [14:26] Vulpes: heh, maybe I should subtract biinfiles from eachother & see what the difference is
  162. [14:27] Jester01: I got the nan here locally
  163. [14:27] Jester01: turns out it only happens if you are NOT optimizing
  164. [14:27] neuro_sys (~neuro@unaffiliated/neurosys/x-283974) joined the chat room.
  165. [14:28] Vulpes: if I'm not optimizing, I'll get a NAN written to my float variables in the .c ?
  166. [14:28] Jester01: strangely enough they are all right in memory
  167. [14:34] Jester01: mmh, you get FPU stack overflow
  168. [14:37] Fabba (unknown@unaffiliated/fabba) joined the chat room.
  169. [14:37] ThFabba (unknown@unaffiliated/fabba) left the chat room. (Disconnected by services)
  170. [14:37] Fabba is now known as ThFabba.
  171. [14:37] Vulpes: hmm, ok.
  172. [14:37] Vulpes: so stackoverflow occurs when I'm just pumping the 1s & 0s into the stack, right?
  173. [14:38] kramaus ([email protected]) joined the chat room.
  174. [14:38] monsterwizard ([email protected]) joined the chat room.
  175. [14:38] Jester01: oh, you fill the fpu stack with 8 numbers, then during printing you pop and push them back
  176. [14:38] nazriel ([email protected]) joined the chat room.
  177. [14:38] Jester01: so they are LEFT on the stack
  178. [14:39] Vulpes: therefore ... I think I'll try unloading the fpu stack first, but by how much? how do I access the fpu status register?
  179. [14:39] Jester01: nonono
  180. [14:39] Jester01: in this case you DO know you have put 8 values on the stack
  181. [14:39] Jester01: but you LEAVE them there!
  182. [14:39] Vulpes: yes
  183. [14:39] Jester01: that's bad.
  184. [14:39] Vulpes: I'm supposed to clear it all out save st0?
  185. [14:39] Jester01: test_floats puts the values there, it should also remove them
  186. [14:39] • Jester01 tries
  187. [14:40] Jester01: erm
  188. [14:40] Jester01: except for 1
  189. [14:40] Jester01: the function return value ;)
  190. [14:40] Jester01: yay that worked
  191. [14:41] • Vulpes gets excited
  192. [14:41] Jester01: I added 7 fstp st0 lines to the end of test_floats
  193. [14:42] Jester01: nevertheless, your print_fpstack should correctly handle the stack size
  194. [14:42] Vulpes: YES!!
  195. [14:42] Vulpes: thank you so much
Advertisement
Add Comment
Please, Sign In to add comment
Public Pastes
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!