[patch] simulate_llsc

[patch] simulate_llsc

[Yoichi Yuasa]

Hi Ralf,

I found a problem in traps.c .
The simulate_llsc does not have return values.

Please apply these patches.

Yoichi

for v2.4
--- ./arch/mips/kernel/traps.c.orig     Mon Apr 21 10:56:54 2003
+++ ./arch/mips/kernel/traps.c  Thu Apr 24 16:25:57 2003
@@ -521,11 +521,11 @@
 
        if ((opcode & OPCODE) == LL) {
                simulate_ll(regs, opcode);
-               return;
+               return 0;
        }
        if ((opcode & OPCODE) == SC) {
                simulate_sc(regs, opcode);
-               return;
+               return 0;
        }
 }

for v2.5
--- ./arch/mips/kernel/traps.c.orig     Mon Apr 21 10:59:49 2003
+++ ./arch/mips/kernel/traps.c  Thu Apr 24 16:14:14 2003
@@ -516,11 +516,11 @@
 
        if ((opcode & OPCODE) == LL) {
                simulate_ll(regs, opcode);
-               return;
+               return 0;
        }
        if ((opcode & OPCODE) == SC) {
                simulate_sc(regs, opcode);
-               return;
+               return 0;
        }
 }

Re: [patch] simulate_llsc

[Ralf Baechle]

On Thu, Apr 24, 2003 at 04:27:49PM +0900, Yoichi Yuasa wrote:

> I found a problem in traps.c .
> The simulate_llsc does not have return values.
> 
> Please apply these patches.

Thanks, applied.

  Ralf

[patch] wait instruction on vr4181

[Steven Seeger]

[-- Attachment #1: Type: text/plain, Size: 106 bytes --]

The VR4181 does work with the wait instruction. This is a patch for that.
(arch/mips/cpu-probe.c)

Steve


[-- Attachment #2: patch-cpu-probe.c --]
[-- Type: application/octet-stream, Size: 323 bytes --]

--- linux-2.4.21.orig/arch/mips/kernel/cpu-probe.c	2003-04-24 05:32:21.000000000 -0700
+++ linux-2.4.21/arch/mips/kernel/cpu-probe.c	2003-04-24 05:26:42.000000000 -0700
@@ -69,6 +69,7 @@
 	case CPU_NEVADA:
 	case CPU_RM7000:
 	case CPU_TX49XX:
+	case CPU_VR4181:
 	case CPU_4KC:
 	case CPU_4KEC:
 	case CPU_4KSC:

[patch] wait instruction on vr4181

[Steven Seeger]

[-- Attachment #1: Type: text/plain, Size: 106 bytes --]

The VR4181 does work with the wait instruction. This is a patch for that.
(arch/mips/cpu-probe.c)

Steve


[-- Attachment #2: patch-cpu-probe.c --]
[-- Type: application/octet-stream, Size: 323 bytes --]

--- linux-2.4.21.orig/arch/mips/kernel/cpu-probe.c	2003-04-24 05:32:21.000000000 -0700
+++ linux-2.4.21/arch/mips/kernel/cpu-probe.c	2003-04-24 05:26:42.000000000 -0700
@@ -69,6 +69,7 @@
 	case CPU_NEVADA:
 	case CPU_RM7000:
 	case CPU_TX49XX:
+	case CPU_VR4181:
 	case CPU_4KC:
 	case CPU_4KEC:
 	case CPU_4KSC:

Re: [patch] wait instruction on vr4181

[Jun Sun]

On Thu, Apr 24, 2003 at 07:10:28AM -0700, Steven Seeger wrote:
> The VR4181 does work with the wait instruction. This is a patch for that.
> (arch/mips/cpu-probe.c)
> 
> Steve
> 

The CPU mannual (v1.1 draft) I have for Vr4181 does not say about "wait"
instruction.  How do you know it has "wait"?

I will ask NEC for confirmation.


Jun

RE: [patch] wait instruction on vr4181

[Steven Seeger]

Oops. I guess I'm tired today. We should use "standby" for our wait
instruction.

I will make a new patch.

Steve

>-----Original Message-----
>From: Jun Sun [mailto:jsun@mvista.com]
>Sent: Thursday, April 24, 2003 9:34 AM
>To: Steven Seeger
>Cc: linux-mips@linux-mips.org; jsun@mvista.com
>Subject: Re: [patch] wait instruction on vr4181
>
>
>
>On Thu, Apr 24, 2003 at 07:10:28AM -0700, Steven Seeger wrote:
>> The VR4181 does work with the wait instruction. This is a
>patch for that.
>> (arch/mips/cpu-probe.c)
>>
>> Steve
>>
>
>The CPU mannual (v1.1 draft) I have for Vr4181 does not say
>about "wait"
>instruction.  How do you know it has "wait"?
>
>I will ask NEC for confirmation.
>
>
>Jun
>

RE: [patch] wait instruction on vr4181

[Steven Seeger]

Oops. I guess I'm tired today. We should use "standby" for our wait
instruction.

I will make a new patch.

Steve

>-----Original Message-----
>From: Jun Sun [mailto:jsun@mvista.com]
>Sent: Thursday, April 24, 2003 9:34 AM
>To: Steven Seeger
>Cc: linux-mips@linux-mips.org; jsun@mvista.com
>Subject: Re: [patch] wait instruction on vr4181
>
>
>
>On Thu, Apr 24, 2003 at 07:10:28AM -0700, Steven Seeger wrote:
>> The VR4181 does work with the wait instruction. This is a
>patch for that.
>> (arch/mips/cpu-probe.c)
>>
>> Steve
>>
>
>The CPU mannual (v1.1 draft) I have for Vr4181 does not say
>about "wait"
>instruction.  How do you know it has "wait"?
>
>I will ask NEC for confirmation.
>
>
>Jun
>

RE: [patch] wait instruction on vr4181

[Steven Seeger]

Hey, when I try to patch in standby, it says that opcode isn't available for
the R6000. Why does arch/mips/Makefile use -mcpu=r4600 for the VR41XX?

Steve

RE: [patch] wait instruction on vr4181

[Steven Seeger]

Hey, when I try to patch in standby, it says that opcode isn't available for
the R6000. Why does arch/mips/Makefile use -mcpu=r4600 for the VR41XX?

Steve

Re: [patch] wait instruction on vr4181

[Jun Sun]

On Thu, Apr 24, 2003 at 11:09:26AM -0700, Steven Seeger wrote:
> Hey, when I try to patch in standby, it says that opcode isn't available for
> the R6000. Why does arch/mips/Makefile use -mcpu=r4600 for the VR41XX?
>

"standby" opcode is probably only recognized by using "-mcpu=r4100". 

I am not sure what impact would it be to change the flag.

Another way to work around is to use the binary value for instruction.

Jun

RE: [patch] wait instruction on vr4181

[Maciej W. Rozycki]

On Thu, 24 Apr 2003, Steven Seeger wrote:

> Hey, when I try to patch in standby, it says that opcode isn't available for
> the R6000. Why does arch/mips/Makefile use -mcpu=r4600 for the VR41XX?

 I fear you need to handcode the instruction until gas supports some sort
of a CPU override, similar to ".set mips".  Using "-mcpu" or "-march" 
won't help as the code wants to be compiled regardless of the CPU
selection.  Currently gas supports the opcode for VR4100, VR4111 and
VR4120, but there is no way to select any of them except on the command
line. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: [patch] wait instruction on vr4181

[Thiemo Seufer]

Maciej W. Rozycki wrote:
> On Thu, 24 Apr 2003, Steven Seeger wrote:
> 
> > Hey, when I try to patch in standby, it says that opcode isn't available for
> > the R6000. Why does arch/mips/Makefile use -mcpu=r4600 for the VR41XX?

Probably because it is the closest match for very old toolchains.

>  I fear you need to handcode the instruction until gas supports some sort
> of a CPU override, similar to ".set mips".  Using "-mcpu" or "-march" 
> won't help as the code wants to be compiled regardless of the CPU
> selection.  Currently gas supports the opcode for VR4100, VR4111 and
> VR4120, but there is no way to select any of them except on the command
> line. 

Huh? Current CVS GAS has command line options for
-march=vr{4100,4111,4120,4130,4181,4300}, and even rather old
GAS versions have -4100. Current CVS gcc supports
-march=vr{4100,4111,4120,4300}.


Thiemo

Re: [patch] wait instruction on vr4181

[Thiemo Seufer]

Thiemo Seufer wrote:
> Maciej W. Rozycki wrote:
> > On Thu, 24 Apr 2003, Steven Seeger wrote:
> > 
> > > Hey, when I try to patch in standby, it says that opcode isn't available for
> > > the R6000. Why does arch/mips/Makefile use -mcpu=r4600 for the VR41XX?
> 
> Probably because it is the closest match for very old toolchains.
> 
> >  I fear you need to handcode the instruction until gas supports some sort
> > of a CPU override, similar to ".set mips".  Using "-mcpu" or "-march" 
> > won't help as the code wants to be compiled regardless of the CPU
> > selection.  Currently gas supports the opcode for VR4100, VR4111 and
> > VR4120, but there is no way to select any of them except on the command
                                                      ^^^^^^
I have missed this word, but I still can't see how it is relevant to
the problem. AFAICS adding '-Wa,-4100' to CFLAGS should solve it,
even in the case of a very old gcc.

> > line. 
> 
> Huh? Current CVS GAS has command line options for
> -march=vr{4100,4111,4120,4130,4181,4300}, and even rather old
> GAS versions have -4100. Current CVS gcc supports
> -march=vr{4100,4111,4120,4300}.


Thiemo

[patch] new wait instruction for vr4181

[Steven Seeger]

[-- Attachment #1: Type: text/plain, Size: 104 bytes --]

I think I figured this out. Could someone look at this and tell me if I did
it right?

Thanks.

-Steve


[-- Attachment #2: cpu-probe.c.diff --]
[-- Type: application/octet-stream, Size: 687 bytes --]

--- /root/vr/new/linux-2.4.21.orig/arch/mips/kernel/cpu-probe.c	2003-04-24 05:32:21.000000000 -0700
+++ ./cpu-probe.c	2003-04-24 14:16:35.000000000 -0700
@@ -34,6 +34,16 @@
 		".set\tmips0");
 }
 
+#ifdef CONFIG_VR4181
+static void vr4181_wait(void)
+{
+   __asm__(".set\tnoreorder\n"
+	   ".int 0x42000021\n"
+	   ".set\treorder\n");
+   
+}
+#endif
+  
 void au1k_wait(void)
 {
 #ifdef CONFIG_PM
@@ -83,6 +93,12 @@
 		cpu_wait = au1k_wait;
 		printk(" available.\n");
 		break;
+#ifdef CONFIG_VR4181
+	 case CPU_VR4181:
+	   cpu_wait = vr4181_wait;
+	   printk(" available\n");
+	   break;
+#endif	   
 	default:
 		printk(" unavailable.\n");
 		break;

[patch] new wait instruction for vr4181

[Steven Seeger]

[-- Attachment #1: Type: text/plain, Size: 104 bytes --]

I think I figured this out. Could someone look at this and tell me if I did
it right?

Thanks.

-Steve


[-- Attachment #2: cpu-probe.c.diff --]
[-- Type: application/octet-stream, Size: 687 bytes --]

--- /root/vr/new/linux-2.4.21.orig/arch/mips/kernel/cpu-probe.c	2003-04-24 05:32:21.000000000 -0700
+++ ./cpu-probe.c	2003-04-24 14:16:35.000000000 -0700
@@ -34,6 +34,16 @@
 		".set\tmips0");
 }
 
+#ifdef CONFIG_VR4181
+static void vr4181_wait(void)
+{
+   __asm__(".set\tnoreorder\n"
+	   ".int 0x42000021\n"
+	   ".set\treorder\n");
+   
+}
+#endif
+  
 void au1k_wait(void)
 {
 #ifdef CONFIG_PM
@@ -83,6 +93,12 @@
 		cpu_wait = au1k_wait;
 		printk(" available.\n");
 		break;
+#ifdef CONFIG_VR4181
+	 case CPU_VR4181:
+	   cpu_wait = vr4181_wait;
+	   printk(" available\n");
+	   break;
+#endif	   
 	default:
 		printk(" unavailable.\n");
 		break;

Re: [patch] new wait instruction for vr4181

[Thiemo Seufer]

Steven Seeger wrote:
> I think I figured this out. Could someone look at this and tell me if I did
> it right?
> 
> Thanks.
> 
> -Steve
> 

> --- /root/vr/new/linux-2.4.21.orig/arch/mips/kernel/cpu-probe.c	2003-04-24 05:32:21.000000000 -0700
> +++ ./cpu-probe.c	2003-04-24 14:16:35.000000000 -0700
> @@ -34,6 +34,16 @@
>  		".set\tmips0");
>  }
>  
> +#ifdef CONFIG_VR4181
> +static void vr4181_wait(void)

The GAS source claims all vr41xx have 'standby', so you may probably
rename the function and enable it for all CONFIG_VR41*.

> +{
> +   __asm__(".set\tnoreorder\n"

A \t at the end like
__asm__(".set\tnoreorder\n\t"
gives better formatting, and the last line doesn't need a \n.

> +	   ".int 0x42000021\n"

.word is probably more usual.


Thiemo

Re: [patch] wait instruction on vr4181

[Maciej W. Rozycki]

On Thu, 24 Apr 2003, Thiemo Seufer wrote:

> I have missed this word, but I still can't see how it is relevant to
> the problem. AFAICS adding '-Wa,-4100' to CFLAGS should solve it,
> even in the case of a very old gcc.

 But if I try to build for anything else, say for R3k or MIPS64, there
will be "-mcpu=r3000" or "-mcpu=r4600" passed and an assembly will fail as
the "standby" instruction won't magically disappear.  That's why
r4k_wait() and au1k_wait() use ".set mips3" for "wait". 

 BTW, "-m4100" and friends are deprecated and their interaction with
"-mcpu=", "-march=" and "-mtune=" is unobvious; I have no idea why they
haven't been banished from the trunk, yet.

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: [patch] wait instruction on vr4181

[Thiemo Seufer]

Maciej W. Rozycki wrote:
> On Thu, 24 Apr 2003, Thiemo Seufer wrote:
> 
> > I have missed this word, but I still can't see how it is relevant to
> > the problem. AFAICS adding '-Wa,-4100' to CFLAGS should solve it,
> > even in the case of a very old gcc.
> 
>  But if I try to build for anything else, say for R3k or MIPS64, there
> will be "-mcpu=r3000" or "-mcpu=r4600" passed and an assembly will fail as
> the "standby" instruction won't magically disappear.  That's why
> r4k_wait() and au1k_wait() use ".set mips3" for "wait". 

The relevant function was #ifdef'ed.

>  BTW, "-m4100" and friends are deprecated and their interaction with
> "-mcpu=", "-march=" and "-mtune=" is unobvious; I have no idea why they
> haven't been banished from the trunk, yet.

True, but as long as people prefer to combine outdated compilers
with the latest binutils this sort of problems will remain.


Thiemo

Re: [patch] wait instruction on vr4181

[Maciej W. Rozycki]

On Fri, 25 Apr 2003, Thiemo Seufer wrote:

> >  But if I try to build for anything else, say for R3k or MIPS64, there
> > will be "-mcpu=r3000" or "-mcpu=r4600" passed and an assembly will fail as
> > the "standby" instruction won't magically disappear.  That's why
> > r4k_wait() and au1k_wait() use ".set mips3" for "wait". 
> 
> The relevant function was #ifdef'ed.

 Whis clutters the source and shouldn't be there IMO.

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: [patch] new wait instruction for vr4181

[Maciej W. Rozycki]

On Thu, 24 Apr 2003, Steven Seeger wrote:

> I think I figured this out. Could someone look at this and tell me if I did
> it right?

 The ifdefs are unnecessary and clutter the code and you should add a
comment near the handcoded instruction, explaining what it is and possibly
sticking a "FIXME" to it.

 BTW, please try to send patches inline in the future. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +