Fix v3 GeodeLX stack and global variable pointer corruption. We had a jump instead of a call to stage1_main in geodelx/stage0.S. That means all accesses to bist and init_detected were off by 8 bytes and collided with accesses to the global variable pointer.
Found during my cleanup runs.
Signed-off-by: Carl-Daniel Hailfinger c-d.hailfinger.devel.2006@gmx.net Acked-by: Carl-Daniel Hailfinger c-d.hailfinger.devel.2006@gmx.net
Index: corebootv3-arch_x86_cleanup/arch/x86/geodelx/stage0.S =================================================================== --- corebootv3-arch_x86_cleanup/arch/x86/geodelx/stage0.S (Revision 907) +++ corebootv3-arch_x86_cleanup/arch/x86/geodelx/stage0.S (Arbeitskopie) @@ -376,7 +376,7 @@ pushl $0 /* First parameter: bist */ pushl %eax - jmp stage1_main + call stage1_main /* We will not go back. */
/* Reset vector. */
Carl-Daniel Hailfinger wrote:
Fix v3 GeodeLX stack and global variable pointer corruption. We had a jump instead of a call to stage1_main in geodelx/stage0.S. That means all accesses to bist and init_detected were off by 8 bytes and collided with accesses to the global variable pointer.
Can you explain what, if any, effect this bug had or could have had in practice?
//Peter
On 08.10.2008 02:29, Peter Stuge wrote:
Carl-Daniel Hailfinger wrote:
Fix v3 GeodeLX stack and global variable pointer corruption. We had a jump instead of a call to stage1_main in geodelx/stage0.S. That means all accesses to bist and init_detected were off by 8 bytes and collided with accesses to the global variable pointer.
Can you explain what, if any, effect this bug had or could have had in practice?
Sure. 1.) If gcc had decided to reload bist from stack after initializing the global variable pointer, bist would have been nonzero, an indicator for processor failure. 2.) If gcc had decided to use the stack location of bist as a scratch register (and it probably is free to do so as long as the contents are restored before returning), it would have clobbered the global variable pointer, leading to NULL pointer dereferences. 3.) Any accesses to init_detected would have resulted in accessing 4 bytes above the top of stack (0x87ffc-0x87fff), something the rest of the code deliberately avoids.
Regards, Carl-Daniel
Ron?
I believe that fix is really needed and the may have been the culprit of some real-world unexplained strangeness.
Regards, Carl-Daniel
On 08.10.2008 03:05, Carl-Daniel Hailfinger wrote:
On 08.10.2008 02:29, Peter Stuge wrote:
Carl-Daniel Hailfinger wrote:
Fix v3 GeodeLX stack and global variable pointer corruption. We had a jump instead of a call to stage1_main in geodelx/stage0.S. That means all accesses to bist and init_detected were off by 8 bytes and collided with accesses to the global variable pointer.
Can you explain what, if any, effect this bug had or could have had in practice?
Sure. 1.) If gcc had decided to reload bist from stack after initializing the global variable pointer, bist would have been nonzero, an indicator for processor failure. 2.) If gcc had decided to use the stack location of bist as a scratch register (and it probably is free to do so as long as the contents are restored before returning), it would have clobbered the global variable pointer, leading to NULL pointer dereferences. 3.) Any accesses to init_detected would have resulted in accessing 4 bytes above the top of stack (0x87ffc-0x87fff), something the rest of the code deliberately avoids.
Regards, Carl-Daniel
On Wed, Oct 8, 2008 at 8:56 AM, Carl-Daniel Hailfinger c-d.hailfinger.devel.2006@gmx.net wrote:
Ron?
I believe that fix is really needed and the may have been the culprit of some real-world unexplained strangeness.
I see it as a very fine catch. I did not see a problem in practice, but wow! what a nice find.
ron
On 08.10.2008 18:00, ron minnich wrote:
On Wed, Oct 8, 2008 at 8:56 AM, Carl-Daniel Hailfinger c-d.hailfinger.devel.2006@gmx.net wrote:
Ron?
I believe that fix is really needed and the may have been the culprit of some real-world unexplained strangeness.
I see it as a very fine catch. I did not see a problem in practice, but wow! what a nice find.
Thanks, r909.
Regards, Carl-Daniel
Carl-Daniel Hailfinger wrote:
Can you explain what, if any, effect this bug had or could have had in practice?
Sure. 1.) If gcc had decided to reload bist from stack after initializing the global variable pointer, bist would have been nonzero, an indicator for processor failure. 2.) If gcc had decided to use the stack location of bist as a scratch register (and it probably is free to do so as long as the contents are restored before returning), it would have clobbered the global variable pointer, leading to NULL pointer dereferences. 3.) Any accesses to init_detected would have resulted in accessing 4 bytes above the top of stack (0x87ffc-0x87fff), something the rest of the code deliberately avoids.
Thanks! If you add the above to the commit message I say:
Acked-by: Peter Stuge peter@stuge.se
-
Carl-Daniel Hailfinger -
Peter Stuge -
ron minnich