[PATCH] fix PPC floating point debug

[PATCH] fix PPC floating point debug

[Stefani Seibold]

The PPC architecture is unable to debug applications using hardware
floating point, because it would not save the floating point registers.

After returning from the debugger, the contents of register was
modified. This patch fix this bug.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 traps.c |    6 ++++++
 1 file changed, 6 insertions(+)

--- linux-2.6.32-rc5/arch/powerpc/kernel/traps.c.orig	2009-11-27 09:47:37.989943124 +0100
+++ linux-2.6.32-rc5/arch/powerpc/kernel/traps.c	2009-11-27 09:47:41.088330825 +0100
@@ -559,6 +559,8 @@ void instruction_breakpoint_exception(st
 		return;
 	if (debugger_iabr_match(regs))
 		return;
+	if (regs->msr & MSR_FP)
+		giveup_fpu(current);
 	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
 }
 
@@ -577,6 +579,8 @@ void __kprobes single_step_exception(str
 	if (debugger_sstep(regs))
 		return;
 
+	if (regs->msr & MSR_FP)
+		giveup_fpu(current);
 	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
 }
 
@@ -834,6 +838,8 @@ void __kprobes program_check_exception(s
 			regs->nip += 4;
 			return;
 		}
+		if (regs->msr & MSR_FP)
+			giveup_fpu(current);
 		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
 		return;
 	}

Re: [PATCH] fix PPC floating point debug

[Benjamin Herrenschmidt]

On Fri, 2009-11-27 at 09:59 +0100, Stefani Seibold wrote:
> The PPC architecture is unable to debug applications using hardware
> floating point, because it would not save the floating point registers.
> 
> After returning from the debugger, the contents of register was
> modified. This patch fix this bug.

I'm not sure what problem you are trying to fix... debugging FP apps
works just fine afaik. You don't need to flush the FP state into the
thread when delivering the SIGTRAP. If you have a signal handler, that
will be done for you by the signal code before laying out the signal
frame. If you are using ptrace, you should be using the appropriate
ptrace calls to retrieve the FP state and they should do the right thing
to.

Do you have a precise example that doesn't work and which is fixed by
your patch ?

Cheers,
Ben.

> Signed-off-by: Stefani Seibold <stefani@seibold.net>
> ---
>  traps.c |    6 ++++++
>  1 file changed, 6 insertions(+)
> 
> --- linux-2.6.32-rc5/arch/powerpc/kernel/traps.c.orig	2009-11-27 09:47:37.989943124 +0100
> +++ linux-2.6.32-rc5/arch/powerpc/kernel/traps.c	2009-11-27 09:47:41.088330825 +0100
> @@ -559,6 +559,8 @@ void instruction_breakpoint_exception(st
>  		return;
>  	if (debugger_iabr_match(regs))
>  		return;
> +	if (regs->msr & MSR_FP)
> +		giveup_fpu(current);
>  	_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
>  }
>  
> @@ -577,6 +579,8 @@ void __kprobes single_step_exception(str
>  	if (debugger_sstep(regs))
>  		return;
>  
> +	if (regs->msr & MSR_FP)
> +		giveup_fpu(current);
>  	_exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
>  }
>  
> @@ -834,6 +838,8 @@ void __kprobes program_check_exception(s
>  			regs->nip += 4;
>  			return;
>  		}
> +		if (regs->msr & MSR_FP)
> +			giveup_fpu(current);
>  		_exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
>  		return;
>  	}
> 
> 
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev@lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [PATCH] fix PPC floating point debug

[Stefani Seibold]

Am Freitag, den 27.11.2009, 22:01 +1100 schrieb Benjamin Herrenschmidt: 
> On Fri, 2009-11-27 at 09:59 +0100, Stefani Seibold wrote:
> > The PPC architecture is unable to debug applications using hardware
> > floating point, because it would not save the floating point registers.
> > 
> > After returning from the debugger, the contents of register was
> > modified. This patch fix this bug.
> 
> I'm not sure what problem you are trying to fix... debugging FP apps
> works just fine afaik. You don't need to flush the FP state into the
> thread when delivering the SIGTRAP. If you have a signal handler, that
> will be done for you by the signal code before laying out the signal
> frame. If you are using ptrace, you should be using the appropriate
> ptrace calls to retrieve the FP state and they should do the right thing
> to.
> 

Aeh... Forget it! Thank you for the support.

Stefani