Thread: __copy_user bug

Brought to you by: kkojima, lethal

linuxsh-dev

__copy_user bug

From: Kristoffer E. <kri...@gm...> - 2007-11-01 10:38:39

Greetings,

Haven't had any feedback so far. I would want that the __copy_user patch got reverted since it obviously doesn't work on atleast SH7709A.
I guess we could ifdef it, but will just make it ugly.
I stand little chance of fixing it myself without Stuarts help.

Best wishes
Kristoffer

Re: __copy_user bug

From: Paul M. <le...@li...> - 2007-11-01 10:45:36

On Thu, Nov 01, 2007 at 11:38:44AM -0700, Kristoffer Ericson wrote:
> Haven't had any feedback so far. I would want that the __copy_user
> patch got reverted since it obviously doesn't work on atleast SH7709A.
> I guess we could ifdef it, but will just make it ugly.  I stand little
> chance of fixing it myself without Stuarts help.
> 
I made some hacks to Stuart's version, so it's possible it was
inadvertently broken at that point. I'll hack together some test code and
see if I can reproduce the failure. Barring that, I'll revert it.

Re: __copy_user bug

From: Stuart M. <stu...@st...> - 2007-11-01 19:56:13

Looking at the code, there is a movca.l in there, which IIRC isn't
available on an SH3.

Although given that, I don't understand how this even builds.

So try this:

Index: linux-2.6.23.1-stm/arch/sh/mm/copy_page.S
===================================================================
--- linux-2.6.23.1-stm.orig/arch/sh/mm/copy_page.S
+++ linux-2.6.23.1-stm/arch/sh/mm/copy_page.S
@@ -256,7 +256,11 @@ EX(	mov.l	@r5+,r8		)
 EX(	mov.l	@r5+,r9		)
 EX(	mov.l	@r5+,r10	)
 EX(	mov.l	@r5+,r11	)
+#if defined(CONFIG_CPU_SH4)
 EX(	movca.l	r0,@r4		)
+#else
+EX(	mov.l	r0,@r4		)
+#endif
 	add	#-32, r6
 EX(	mov.l	r1,@(4,r4)	)
 	mov	#32, r0


Kristoffer Ericson wrote:
> Greetings,
>
> Haven't had any feedback so far. I would want that the __copy_user patch got reverted since it obviously doesn't work on atleast SH7709A.
> I guess we could ifdef it, but will just make it ugly.
> I stand little chance of fixing it myself without Stuarts help.
>
> Best wishes
> Kristoffer
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems?  Stop.
> Now Search log events and configuration files using AJAX and a browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
> _______________________________________________
> linuxsh-dev mailing list
> lin...@li...
> https://lists.sourceforge.net/lists/listinfo/linuxsh-dev

Re: __copy_user bug

From: Kristoffer E. <kri...@gm...> - 2007-11-01 20:45:04

Seems like you are quite correct Stuart. Applying this patch makes everything work fine, removing patch produces bug again.
Nice catch!


On Thu, 01 Nov 2007 19:55:57 +0000
Stuart MENEFY <stu...@st...> wrote:

> Looking at the code, there is a movca.l in there, which IIRC isn't
> available on an SH3.
> 
> Although given that, I don't understand how this even builds.
Doesn't produce any warnings/errors at all.

> 
> So try this:
> 
> Index: linux-2.6.23.1-stm/arch/sh/mm/copy_page.S
> ===================================================================
> --- linux-2.6.23.1-stm.orig/arch/sh/mm/copy_page.S
> +++ linux-2.6.23.1-stm/arch/sh/mm/copy_page.S
> @@ -256,7 +256,11 @@ EX(	mov.l	@r5+,r8		)
>  EX(	mov.l	@r5+,r9		)
>  EX(	mov.l	@r5+,r10	)
>  EX(	mov.l	@r5+,r11	)
> +#if defined(CONFIG_CPU_SH4)
>  EX(	movca.l	r0,@r4		)
> +#else
> +EX(	mov.l	r0,@r4		)
> +#endif
>  	add	#-32, r6
>  EX(	mov.l	r1,@(4,r4)	)
>  	mov	#32, r0
> 
> 
> Kristoffer Ericson wrote:
> > Greetings,
> >
> > Haven't had any feedback so far. I would want that the __copy_user patch got reverted since it obviously doesn't work on atleast SH7709A.
> > I guess we could ifdef it, but will just make it ugly.
> > I stand little chance of fixing it myself without Stuarts help.
> >
> > Best wishes
> > Kristoffer
> >
> > -------------------------------------------------------------------------
> > This SF.net email is sponsored by: Splunk Inc.
> > Still grepping through log files to find problems?  Stop.
> > Now Search log events and configuration files using AJAX and a browser.
> > Download your FREE copy of Splunk now >> http://get.splunk.com/
> > _______________________________________________
> > linuxsh-dev mailing list
> > lin...@li...
> > https://lists.sourceforge.net/lists/listinfo/linuxsh-dev
>

Re: __copy_user bug

From: Paul M. <le...@li...> - 2007-11-02 01:36:42

On Thu, Nov 01, 2007 at 07:55:57PM +0000, Stuart MENEFY wrote:
> Looking at the code, there is a movca.l in there, which IIRC isn't
> available on an SH3.
> 
> Although given that, I don't understand how this even builds.
> 
In order for it to blow up the ISA tuning has to be just right, or else
any supported opcode won't complain. This is part of why the ISA
selection in arch/sh/Makefile is such a pain, especially as the various
toolchains make it a total grab-bag with regards to what flags can be
supported and which ones can't.

Initially it was helpful on SH-2/SH-2A for finding all of the register
bank accessors, as otherwise they would all build in successfully and
just generate some fairly interesting bugs.

Some places manage to elude the ISA tuning completely as well by just
hardcoding the opcodes and operands in hex due to certain toolchains not
undertanding the opcode mnemonics at all. This happens with the DSP state
save/restore, for example.

It's something we can slowly fix up, but there are unfortunately far too
many varied toolchains out in the wild with utterl conflicting behaviour.