[PATCH] fix local_irq_save()/local_irq_restore() when CONFIG_CPU_MIPSR2 & CONFIG_IRQ_CPU

[PATCH] fix local_irq_save()/local_irq_restore() when CONFIG_CPU_MIPSR2 & CONFIG_IRQ_CPU

[Maxime Bizon]

Hello all,

I was unable to get my R4KECr2 board working with CONFIG_CPU_MIPSR2 &
CONFIG_IRQ_CPU, irq delivery is not working.

local_irq_restore() logic is to check that "flags" is non zero, and
enable irq accordingly.

But "flags" comes directly from di, which according to mips instruction
set, saves whole status content, not just IE bit.

Attached patch to fix this.


Signed-off-by: Maxime Bizon <mbizon@freebox.fr>

--- linux.git/include/asm-mips/interrupt.h.old	2005-12-20 06:20:44.000000000 +0100
+++ linux.git/include/asm-mips/interrupt.h	2005-12-20 06:21:02.000000000 +0100
@@ -93,6 +93,7 @@
 	"	.set	noat						\n"
 #ifdef CONFIG_CPU_MIPSR2
 	"	di	\\result					\n"
+	"	andi	\\result, 1					\n"
 #else
 	"	mfc0	\\result, $12					\n"
 	"	ori	$1, \\result, 1					\n"



Thanks,

-- 
Maxime

Re: [PATCH] fix local_irq_save()/local_irq_restore() when CONFIG_CPU_MIPSR2 & CONFIG_IRQ_CPU

[Maciej W. Rozycki]

On Tue, 20 Dec 2005, Maxime Bizon wrote:

> But "flags" comes directly from di, which according to mips instruction
> set, saves whole status content, not just IE bit.
> 
> Attached patch to fix this.

 Or alternatively something like this: 
"http://www.linux-mips.org/cgi-bin/mesg.cgi?a=linux-mips&i=Pine.LNX.4.61L.0507121626370.14155%40blysk.ds.pg.gda.pl", 
for consistency with the other variants.  It looks like we are the only 
two who care...

  Maciej

Re: [PATCH] fix local_irq_save()/local_irq_restore() when CONFIG_CPU_MIPSR2 & CONFIG_IRQ_CPU

[Ralf Baechle]

On Tue, Dec 20, 2005 at 06:32:19AM +0100, Maxime Bizon wrote:

> 
> I was unable to get my R4KECr2 board working with CONFIG_CPU_MIPSR2 &
> CONFIG_IRQ_CPU, irq delivery is not working.
> 
> local_irq_restore() logic is to check that "flags" is non zero, and
> enable irq accordingly.
> 
> But "flags" comes directly from di, which according to mips instruction
> set, saves whole status content, not just IE bit.

There are less local_irq_save than local_irq_restore calls, so is the
preferable place for the fix.  Applied,

  Ralf