Re: [MIPS] Fix handling of trap and breakpoint instructions

Re: [MIPS] Fix handling of trap and breakpoint instructions

[Atsushi Nemoto]

> Author: Ralf Baechle <ralf@linux-mips.org> Sun Apr 20 16:28:54 2008 +0100
> Commit: 5881bb0de64887a60f7f49922cf73a3b4d40fc01
> Gitweb: http://www.linux-mips.org/g/linux/5881bb0d
> Branch: master

You must drop left shift of this line too.

		if (bcode == (BRK_DIVZERO << 10))

---
Atsushi Nemoto

Re: [MIPS] Fix handling of trap and breakpoint instructions

[Ralf Baechle]

On Mon, Apr 21, 2008 at 10:07:21AM +0900, Atsushi Nemoto wrote:

> > Author: Ralf Baechle <ralf@linux-mips.org> Sun Apr 20 16:28:54 2008 +0100
> > Commit: 5881bb0de64887a60f7f49922cf73a3b4d40fc01
> > Gitweb: http://www.linux-mips.org/g/linux/5881bb0d
> > Branch: master
> 
> You must drop left shift of this line too.
> 
> 		if (bcode == (BRK_DIVZERO << 10))

Sigh, damn code duplication.  Will fix and cleanup.

  Ralf

Re: [MIPS] Fix handling of trap and breakpoint instructions

[David Daney]

Atsushi Nemoto wrote:
>> Author: Ralf Baechle <ralf@linux-mips.org> Sun Apr 20 16:28:54 2008 +0100
>> Commit: 5881bb0de64887a60f7f49922cf73a3b4d40fc01
>> Gitweb: http://www.linux-mips.org/g/linux/5881bb0d
>> Branch: master
> 
> You must drop left shift of this line too.
> 
> 		if (bcode == (BRK_DIVZERO << 10))
> 

Note that there has been some confusion about break codes in gas over 
the years.  Ancient versions (I am not sure which) generated different 
break codes than recent versions.

Before changing it make sure that you don't break existing user space code.

One problem (fixed around 2.4.25 or so) was the integer division by zero 
in user space would result in SIGTRAP instead of SIGFPE.   If you change 
the break code handling you should verify that you don't break this.

David Daney

Re: [MIPS] Fix handling of trap and breakpoint instructions

[Ralf Baechle]

On Mon, Apr 21, 2008 at 08:47:52AM -0700, David Daney wrote:

> Note that there has been some confusion about break codes in gas over the 
> years.  Ancient versions (I am not sure which) generated different break 
> codes than recent versions.
>
> Before changing it make sure that you don't break existing user space code.
>
> One problem (fixed around 2.4.25 or so) was the integer division by zero in 
> user space would result in SIGTRAP instead of SIGFPE.   If you change the 
> break code handling you should verify that you don't break this.

Not quite correct.  Very old and recent binutils are bug compatible.  It's
middle age gcc which unfortunately fixed the bug creating a bug ;-)

  Ralf