Problem booting on some HP systems

Problem booting on some HP systems

[Luck, Tony]

Recent kernels are not booting on some HP systems (though
it does boot on others).  James spent time and tracked it
to this patch in the pvops series:

commit 498c5170472ff0c03a29d22dbd33225a0be038f4
Author: Isaku Yamahata <yamahata@valinux.co.jp>
Date:   Mon May 19 22:13:38 2008 +0900

    [IA64] pvops: paravirtualize ivt.S

I did a quick compile with and without this patch applied and
compared the diassembled ivt.o file and spotted two places
where an _FROM_ macro had been used where _TO_ was intended.
See the <<<<< OOPS below:


@@ -1360,26 +1357,26 @@ ENTRY(speculation_vector)
         *
         * cr.imm contains zero_ext(imm21)
         */
-       mov r18=cr.iim
+       MOV_FROM_IIM(r18)
        ;;
-       mov r17=cr.iip
+       MOV_FROM_IIP(r17)
        shl r18=r18,43                  // put sign bit in position (43d-21)
        ;;

-       mov r16=cr.ipsr
+       MOV_FROM_IPSR(p0, r16)
        shr r18=r18,39                  // sign extend (39C-4)
        ;;

        add r17=r17,r18                 // now add the offset
        ;;
-       mov cr.iip=r17
+       MOV_FROM_IIP(r17) <<<<<<< OOPS!
        dep r16=0,r16,41,2              // clear EI
        ;;

-       mov cr.ipsr=r16
+       MOV_FROM_IPSR(p0, r16) <<<<<<<< OOPS!
        ;;

-       rfi                             // and go back
+       RFI
 END(speculation_vector)


I don't know whether these are the cause of the problems on James'
and Willy's machines (could you try swapping these round please?)

-Tony

Re: Problem booting on some HP systems

[James Bottomley]

On Fri, 2008-08-01 at 16:24 -0700, Luck, Tony wrote:
> Recent kernels are not booting on some HP systems (though
> it does boot on others).  James spent time and tracked it
> to this patch in the pvops series:
> 
> commit 498c5170472ff0c03a29d22dbd33225a0be038f4
> Author: Isaku Yamahata <yamahata@valinux.co.jp>
> Date:   Mon May 19 22:13:38 2008 +0900
> 
>     [IA64] pvops: paravirtualize ivt.S
> 
> I did a quick compile with and without this patch applied and
> compared the diassembled ivt.o file and spotted two places
> where an _FROM_ macro had been used where _TO_ was intended.
> See the <<<<< OOPS below:
> 
> 
> @@ -1360,26 +1357,26 @@ ENTRY(speculation_vector)
>          *
>          * cr.imm contains zero_ext(imm21)
>          */
> -       mov r18=cr.iim
> +       MOV_FROM_IIM(r18)
>         ;;
> -       mov r17=cr.iip
> +       MOV_FROM_IIP(r17)
>         shl r18=r18,43                  // put sign bit in position (43d-21)
>         ;;
> 
> -       mov r16=cr.ipsr
> +       MOV_FROM_IPSR(p0, r16)
>         shr r18=r18,39                  // sign extend (39C-4)
>         ;;
> 
>         add r17=r17,r18                 // now add the offset
>         ;;
> -       mov cr.iip=r17
> +       MOV_FROM_IIP(r17) <<<<<<< OOPS!
>         dep r16=0,r16,41,2              // clear EI
>         ;;
> 
> -       mov cr.ipsr=r16
> +       MOV_FROM_IPSR(p0, r16) <<<<<<<< OOPS!
>         ;;
> 
> -       rfi                             // and go back
> +       RFI
>  END(speculation_vector)
> 
> 
> I don't know whether these are the cause of the problems on James'
> and Willy's machines (could you try swapping these round please?)

Actually tony traced this to errors in the statements by comparing
disassembled output.  There's two register moves that are now the wrong
way around.  Applying this patch gets my system booting again.

James

---

diff --git a/arch/ia64/kernel/ivt.S b/arch/ia64/kernel/ivt.S
index c39627d..b8750a0 100644
--- a/arch/ia64/kernel/ivt.S
+++ b/arch/ia64/kernel/ivt.S
@@ -1243,11 +1243,11 @@ ENTRY(speculation_vector)
 
 	add r17=r17,r18			// now add the offset
 	;;
-	MOV_FROM_IIP(r17)
+	MOV_TO_IIP(r17, p0)
 	dep r16=0,r16,41,2		// clear EI
 	;;
 
-	MOV_FROM_IPSR(p0, r16)
+	MOV_TO_IPSR(p0, r16, p0)
 	;;
 
 	RFI

Re: Problem booting on some HP systems

[Isaku Yamahata]

On Fri, Aug 01, 2008 at 06:43:08PM -0500, James Bottomley wrote:
> On Fri, 2008-08-01 at 16:24 -0700, Luck, Tony wrote:
> > Recent kernels are not booting on some HP systems (though
> > it does boot on others).  James spent time and tracked it
> > to this patch in the pvops series:
> > 
> > commit 498c5170472ff0c03a29d22dbd33225a0be038f4
> > Author: Isaku Yamahata <yamahata@valinux.co.jp>
> > Date:   Mon May 19 22:13:38 2008 +0900
> > 
> >     [IA64] pvops: paravirtualize ivt.S
> > 
> > I did a quick compile with and without this patch applied and
> > compared the diassembled ivt.o file and spotted two places
> > where an _FROM_ macro had been used where _TO_ was intended.
> > See the <<<<< OOPS below:
> > 
> > 
> > @@ -1360,26 +1357,26 @@ ENTRY(speculation_vector)
> >          *
> >          * cr.imm contains zero_ext(imm21)
> >          */
> > -       mov r18=cr.iim
> > +       MOV_FROM_IIM(r18)
> >         ;;
> > -       mov r17=cr.iip
> > +       MOV_FROM_IIP(r17)
> >         shl r18=r18,43                  // put sign bit in position (43d-21)
> >         ;;
> > 
> > -       mov r16=cr.ipsr
> > +       MOV_FROM_IPSR(p0, r16)
> >         shr r18=r18,39                  // sign extend (39C-4)
> >         ;;
> > 
> >         add r17=r17,r18                 // now add the offset
> >         ;;
> > -       mov cr.iip=r17
> > +       MOV_FROM_IIP(r17) <<<<<<< OOPS!
> >         dep r16=0,r16,41,2              // clear EI
> >         ;;
> > 
> > -       mov cr.ipsr=r16
> > +       MOV_FROM_IPSR(p0, r16) <<<<<<<< OOPS!
> >         ;;
> > 
> > -       rfi                             // and go back
> > +       RFI
> >  END(speculation_vector)
> > 
> > 
> > I don't know whether these are the cause of the problems on James'
> > and Willy's machines (could you try swapping these round please?)
> 
> Actually tony traced this to errors in the statements by comparing
> disassembled output.  There's two register moves that are now the wrong
> way around.  Applying this patch gets my system booting again.
> 
> James
> 
> ---
> 
> diff --git a/arch/ia64/kernel/ivt.S b/arch/ia64/kernel/ivt.S
> index c39627d..b8750a0 100644
> --- a/arch/ia64/kernel/ivt.S
> +++ b/arch/ia64/kernel/ivt.S
> @@ -1243,11 +1243,11 @@ ENTRY(speculation_vector)
>  
>  	add r17=r17,r18			// now add the offset
>  	;;
> -	MOV_FROM_IIP(r17)
> +	MOV_TO_IIP(r17, p0)
>  	dep r16=0,r16,41,2		// clear EI
>  	;;
>  
> -	MOV_FROM_IPSR(p0, r16)
> +	MOV_TO_IPSR(p0, r16, p0)
>  	;;
>  
>  	RFI

Ouch. Sorry for those.
The last argument for those macro should be regular register.
Here is the patch.


From cc61356b6cb83aad5e798c402fe2a323953bb3de Mon Sep 17 00:00:00 2001
From: Isaku Yamahata <yamahata@valinux.co.jp>
Date: Mon, 4 Aug 2008 12:02:28 +0900
Subject: [PATCH] [IA64] pv_ops: fix ivt.S paravirtualization

Recent kernels are not booting on some HP systems (though
it does boot on others).
Two instructions were wrongly paravirtualized such that
an _FROM_ macro had been used where _TO_ was intended
by c/s 498c5170472ff0c03a29d22dbd33225a0be038f4.
This patch fixes them.

Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: "Wilcox, Matthew  R" <matthew.r.wilcox@intel.com>
Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
---
 arch/ia64/kernel/ivt.S |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/ia64/kernel/ivt.S b/arch/ia64/kernel/ivt.S
index c39627d..416a952 100644
--- a/arch/ia64/kernel/ivt.S
+++ b/arch/ia64/kernel/ivt.S
@@ -1243,11 +1243,11 @@ ENTRY(speculation_vector)
 
 	add r17=r17,r18			// now add the offset
 	;;
-	MOV_FROM_IIP(r17)
+	MOV_TO_IIP(r17, r19)
 	dep r16=0,r16,41,2		// clear EI
 	;;
 
-	MOV_FROM_IPSR(p0, r16)
+	MOV_TO_IPSR(p0, r16, r19)
 	;;
 
 	RFI
-- 
1.6.0.rc0.42.g186458

-- 
yamahata