Libcudest - revision history

Dank: /* disassembly */

2011-08-22T22:18:45Z

disassembly

← Older revision		Revision as of 22:18, 22 August 2011
Line 470:		Line 470:

	==disassembly==		==disassembly==
	These ~~disassembly~~ makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.		These disassemblies makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.
	AMD64 ABI:		* AMD64 ABI:
	* Integer arguments via RDI, RSI, RDX, RCX, R8 and R9, then stack		** Integer arguments via RDI, RSI, RDX, RCX, R8 and R9, then stack
	* FP arguments in XMM0..XMM7, then stack		** FP arguments in XMM0..XMM7, then stack
	* Return value in RAX		** Return value in RAX
	* [[libcuda traces]]		** [[libcuda traces]]

	==See Also==		==See Also==

Dank at 07:36, 18 July 2010

2010-07-18T07:36:26Z

← Older revision		Revision as of 07:36, 18 July 2010
Line 1:		Line 1:
			__INDEX__
	Reverse engineering of the [[CUDA]] system. CUDA primarily communicates with the NVIDIA closed-source driver via several dozen undocumented ioctl()s. My open source implementation, libcudest, is located at [http://github.com/dankamongmen/libcudest GitHub]. Sundry utilities for reverse engineering are also within this repository, though recent modifications to [http://kadu.net/~joi/valgrind-mmt.git/ valgrind-mmt] have rather superseded my tools.		Reverse engineering of the [[CUDA]] system. CUDA primarily communicates with the NVIDIA closed-source driver via several dozen undocumented ioctl()s. My open source implementation, libcudest, is located at [http://github.com/dankamongmen/libcudest GitHub]. Sundry utilities for reverse engineering are also within this repository, though recent modifications to [http://kadu.net/~joi/valgrind-mmt.git/ valgrind-mmt] have rather superseded my tools.

Dank: /* disassembly */

2010-07-08T11:07:07Z

disassembly

Show changes

Dank: /* GPU methods */

2010-07-05T10:04:09Z

GPU methods

← Older revision		Revision as of 10:04, 5 July 2010
Line 457:		Line 457:
	\|-		\|-
	\| 0x20800110		\| 0x20800110
	\| 84		\| 0x84
	\|		\|
	* Retrieves device name:		* Retrieves device name:

Dank: /* GPU methods */

2010-07-05T09:57:06Z

GPU methods

← Older revision		Revision as of 09:57, 5 July 2010
Line 457:		Line 457:
	\|-		\|-
	\| 0x20800110		\| 0x20800110
	\| ~~0x84~~		\| 84
	\|		\|
	* Retrieves device name:		* Retrieves device name:
Line 467:		Line 467:
	\|-		\|-
	\|}		\|}

	==disassembly==		==disassembly==
	This disassembly makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.		This disassembly makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.

Dank at 09:09, 5 July 2010

2010-07-05T09:09:45Z

← Older revision		Revision as of 09:09, 5 July 2010
Line 466:		Line 466:
	* 6f46654720656372205354474d303633 == "oFeG ecr STGM063"		* 6f46654720656372205354474d303633 == "oFeG ecr STGM063"
	\|-		\|-
	}		\|}
	==disassembly==		==disassembly==
	This disassembly makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.		This disassembly makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.

Dank at 09:08, 5 July 2010

2010-07-05T09:08:49Z

@@ Line 448: / Line 448: @@
 |-
 |}
+==GPU methods==
 ==disassembly==
 This disassembly makes use of <tt>libcuda.so.195.36.15</tt> (0867d66be617faab3782fa0ba19ec9ba, 7404990 bytes). Symbols were extracted via <tt>objdump -T</tt>.

Dank: /* ioctls */

2010-07-05T09:02:52Z

ioctls

@@ Line 337: / Line 337: @@
 | cuInit
+|
-* GPU method invocation. Second and third words specify the method being called. Fifth and sixth specify the address being passed; seventh and eighth the size thereof.
+* [[#GPU methods|GPU method]] invocation. Second and third words specify the method being called. Fifth and sixth specify the address being passed; seventh and eighth the size thereof.
 Sample inputs:
 <pre>0x7fffffffd310:	3251635025	3251635025	533	0
 x7fffffffd320:	4294955888	32767	132	0</pre>
 * First and second words are *not* always equivalent.
-* Outputs are unchanged (most of the time...)
+* Outputs are usually unchanged, but not always:
 |-
 | 0x2b

Dank: /* ioctls */

2010-07-05T08:31:10Z

ioctls

← Older revision		Revision as of 08:31, 5 July 2010
Line 337:		Line 337:
	\| cuInit		\| cuInit
	\|		\|
	* GPU method invocation~~(?)~~		* GPU method invocation. Second and third words specify the method being called. Fifth and sixth specify the address being passed; seventh and eighth the size thereof.
	~~Inputs~~:		Sample inputs:
	<pre>0x7fffffffd310: 3251635025 3251635025 533 0		<pre>0x7fffffffd310: 3251635025 3251635025 533 0
	0x7fffffffd320: 4294955888 32767 132 0</pre>		0x7fffffffd320: 4294955888 32767 132 0</pre>
	* Outputs are unchanged		* First and second words are not always equivalent.
			* Outputs are unchanged (most of the time...)
	\|-		\|-
	\| 0x2b		\| 0x2b

Dank at 08:22, 5 July 2010

2010-07-05T08:22:22Z

← Older revision		Revision as of 08:22, 5 July 2010
Line 1:		Line 1:
	Reverse engineering of the [[CUDA]] system. CUDA primarily communicates with the NVIDIA closed-source driver via several dozen undocumented ioctl()s. My open source implementation, libcudest, is located at [http://github.com/dankamongmen/libcudest GitHub]. It began as a project for Hyesoon Kim's [[Grad school\|CS4803DGC]] at the Georgia Institute of Technology.		Reverse engineering of the [[CUDA]] system. CUDA primarily communicates with the NVIDIA closed-source driver via several dozen undocumented ioctl()s. My open source implementation, libcudest, is located at [http://github.com/dankamongmen/libcudest GitHub]. Sundry utilities for reverse engineering are also within this repository, though recent modifications to [http://kadu.net/~joi/valgrind-mmt.git/ valgrind-mmt] have rather superseded my tools.

			libcudest began as a project for Hyesoon Kim's [[Grad school\|CS4803DGC]] at the Georgia Institute of Technology.
	==Driver versions==		==Driver versions==
	Newer drivers can be used with older CUDA versions, but the converse is not true. The "CUDA macroversion" listed below is the first CUDA release designed explicitly for use with the listed drivers.		Newer drivers can be used with older CUDA versions, but the converse is not true. The "CUDA macroversion" listed below is the first CUDA release designed explicitly for use with the listed drivers.
Line 335:		Line 337:
	\| cuInit		\| cuInit
	\|		\|
	* Inputs:		* GPU method invocation(?)
			Inputs:
	<pre>0x7fffffffd310: 3251635025 3251635025 533 0		<pre>0x7fffffffd310: 3251635025 3251635025 533 0
	0x7fffffffd320: 4294955888 32767 132 0</pre>		0x7fffffffd320: 4294955888 32767 132 0</pre>
Line 345:		Line 348:
	\| cuInit		\| cuInit
	\|		\|
			* GPU object creation(?)
	\|-		\|-
	\| 0x4d		\| 0x4d