[PATCH 03/15] sparc32: implement proper LEON support in head_32 (after highmem)

[PATCH 03/15] sparc32: implement proper LEON support in head_32 (after highmem)

[Sam Ravnborg]

We use the compatibility property to determine the
sun models. For leon we use psr.impl and ignore the
result of the getprops call.

Include a hack to allow build as the support code
is not yet converted.

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Cc: Daniel Hellstrom <daniel@gaisler.com>
Cc: Konrad Eisele <konrad@gaisler.com>
---
 arch/sparc/kernel/head_32.S |   58 ++++++++++++++++++++++++------------------
 1 files changed, 33 insertions(+), 25 deletions(-)

diff --git a/arch/sparc/kernel/head_32.S b/arch/sparc/kernel/head_32.S
index 5a418d3..f22a729 100644
--- a/arch/sparc/kernel/head_32.S
+++ b/arch/sparc/kernel/head_32.S
@@ -372,37 +372,27 @@ execute_in_high_mem:
 		sethi	%hi(linux_dbvec), %g1
 		st	%o1, [%g1 + %lo(linux_dbvec)]
 
-/* Get the machine type via the mysterious romvec node operations. */
-
-		add	%g7, 0x1c, %l1
-		ld	[%l1], %l0
-		ld	[%l0], %l0
-		call	%l0
-		 or	%g0, %g0, %o0		! next_node(0) = first_node
-		or	%o0, %g0, %g6
-
-		sethi	%hi(cputypvar), %o1	! First node has cpu-arch
-		or	%o1, %lo(cputypvar), %o1
-		sethi	%hi(cputypval), %o2	! information, the string
-		or	%o2, %lo(cputypval), %o2
-		ld	[%l1], %l0		! 'compatible' tells
-		ld	[%l0 + 0xc], %l0	! that we want 'sun4x' where
-		call	%l0			! x is one of 'm', 'd' or 'e'.
-		 nop				! %o2 holds pointer
-						! to a buf where above string
-						! will get stored by the prom.
+		/* Check if this is a LEON CPU.
+		 * Skip getprops call if it is
+		 */
+		srl	%g3, PSR_IMPL_SHIFT, %g3
+		and	%g3, PSR_IMPL_SHIFTED_MASK, %g3
+		cmp	%g3, PSR_IMPL_LEON
+		 bne	get_cputype
 
-#ifdef CONFIG_SPARC_LEON
-	        /* no cpu-type check is needed, it is a SPARC-LEON */
 
+		/* LEON CPU - set boot_cpu_id */
 		sethi	%hi(boot_cpu_id), %g2	! boot-cpu index
 
 #ifdef CONFIG_SMP
 		ldub	[%g2 + %lo(boot_cpu_id)], %g1
 		cmp	%g1, 0xff		! unset means first CPU
+#ifdef CONFIG_SPARC_LEON
+		/* XXX Hack to allow build - remove ifdef later */
 		bne	leon_smp_cpu_startup	! continue only with master
 		 nop
 #endif
+#endif
 		/* Get CPU-ID from most significant 4-bit of ASR17 */
 		rd     %asr17, %g1
 		srl    %g1, 28, %g1
@@ -412,12 +402,30 @@ execute_in_high_mem:
 
 		ba continue_boot
 		 nop
-#endif
+
+/* Get the machine type via the mysterious romvec node operations. */
+get_cputype:
+		add	%g7, 0x1c, %l1
+		ld	[%l1], %l0
+		ld	[%l0], %l0
+		call	%l0
+		 or	%g0, %g0, %o0		! next_node(0) = first_node
+		or	%o0, %g0, %g6
+
+		sethi	%hi(cputypvar), %o1	! First node has cpu-arch
+		or	%o1, %lo(cputypvar), %o1
+		sethi	%hi(cputypval), %o2	! information, the string
+		or	%o2, %lo(cputypval), %o2
+		ld	[%l1], %l0		! 'compatible' tells
+		ld	[%l0 + 0xc], %l0	! that we want 'sun4x' where
+		call	%l0			! x is one of 'm', 'd' or 'e'.
+		 nop				! %o2 holds pointer
+						! to a buf where above string
+						! will get stored by the prom.
 
 /* Check to cputype. We may be booted on a sun4u (64 bit box),
  * and sun4d needs special treatment.
  */
-
 		set	cputypval, %o2
 		ldub	[%o2 + 0x4], %l1
 
@@ -467,9 +475,9 @@ sun4m_init:
 /* This sucks, apparently this makes Vikings call prom panic, will fix later */
 2:
 		rd	%psr, %o1
-		srl	%o1, 28, %o1		! Get a type of the CPU
+		srl	%o1, PSR_IMPL_SHIFT, %o1	! Get a type of the CPU
 
-		subcc	%o1, 4, %g0		! TI: Viking or MicroSPARC
+		subcc	%o1, PSR_IMPL_TI, %g0		! TI: Viking or MicroSPARC
 		be	continue_boot
 		 nop
 
-- 
1.6.0.6

Re: [PATCH 03/15] sparc32: implement proper LEON support in head_32 (after highmem)

[Daniel Hellstrom]

Hello Sam,

I have tested the last patch series [1..15], however LEON can boot. I'm trying to figure out why, this patch removes the cputypval for LEON which makes the architecture detection not be executed. 
Since some things nowdays default to LEON I suggest that the cputypval string defaults to "LEON" as well, or that the check is reintroduced, or that cputypval is set to "LEON" when a LEON is detected. 
The compatible/compatability property is set to LEON on LEON systems. What to you think?


My boot now looks like this:

PROMLIB: Sun Boot Prom Version 0 Revision 0
Linux version 3.4.0-04858-g6211f35 (daniel@daniel) (gcc version 4.4.2 (crosstoo2
bootconsole [earlyprom0] enabled
ARCH: SUN4M
TYPE: Leon3 System-on-a-Chip
Ethernet address: 00:00:7c:cc:01:45
Could not determine SRMMU chip type.
PROM: Halt ...

Daniel


On 05/26/2012 09:20 AM, Sam Ravnborg wrote:
> We use the compatibility property to determine the
> sun models. For leon we use psr.impl and ignore the
> result of the getprops call.
>
> Include a hack to allow build as the support code
> is not yet converted.
>
> Signed-off-by: Sam Ravnborg<sam@ravnborg.org>
> Cc: Daniel Hellstrom<daniel@gaisler.com>
> Cc: Konrad Eisele<konrad@gaisler.com>
> ---
>   arch/sparc/kernel/head_32.S |   58 ++++++++++++++++++++++++------------------
>   1 files changed, 33 insertions(+), 25 deletions(-)
>
> diff --git a/arch/sparc/kernel/head_32.S b/arch/sparc/kernel/head_32.S
> index 5a418d3..f22a729 100644
> --- a/arch/sparc/kernel/head_32.S
> +++ b/arch/sparc/kernel/head_32.S
> @@ -372,37 +372,27 @@ execute_in_high_mem:
>   		sethi	%hi(linux_dbvec), %g1
>   		st	%o1, [%g1 + %lo(linux_dbvec)]
>
> -/* Get the machine type via the mysterious romvec node operations. */
> -
> -		add	%g7, 0x1c, %l1
> -		ld	[%l1], %l0
> -		ld	[%l0], %l0
> -		call	%l0
> -		 or	%g0, %g0, %o0		! next_node(0) = first_node
> -		or	%o0, %g0, %g6
> -
> -		sethi	%hi(cputypvar), %o1	! First node has cpu-arch
> -		or	%o1, %lo(cputypvar), %o1
> -		sethi	%hi(cputypval), %o2	! information, the string
> -		or	%o2, %lo(cputypval), %o2
> -		ld	[%l1], %l0		! 'compatible' tells
> -		ld	[%l0 + 0xc], %l0	! that we want 'sun4x' where
> -		call	%l0			! x is one of 'm', 'd' or 'e'.
> -		 nop				! %o2 holds pointer
> -						! to a buf where above string
> -						! will get stored by the prom.
> +		/* Check if this is a LEON CPU.
> +		 * Skip getprops call if it is
> +		 */
> +		srl	%g3, PSR_IMPL_SHIFT, %g3
> +		and	%g3, PSR_IMPL_SHIFTED_MASK, %g3
> +		cmp	%g3, PSR_IMPL_LEON
> +		 bne	get_cputype
>
> -#ifdef CONFIG_SPARC_LEON
> -	        /* no cpu-type check is needed, it is a SPARC-LEON */
>
> +		/* LEON CPU - set boot_cpu_id */
>   		sethi	%hi(boot_cpu_id), %g2	! boot-cpu index
>
>   #ifdef CONFIG_SMP
>   		ldub	[%g2 + %lo(boot_cpu_id)], %g1
>   		cmp	%g1, 0xff		! unset means first CPU
> +#ifdef CONFIG_SPARC_LEON
> +		/* XXX Hack to allow build - remove ifdef later */
>   		bne	leon_smp_cpu_startup	! continue only with master
>   		 nop
>   #endif
> +#endif
>   		/* Get CPU-ID from most significant 4-bit of ASR17 */
>   		rd     %asr17, %g1
>   		srl    %g1, 28, %g1
> @@ -412,12 +402,30 @@ execute_in_high_mem:
>
>   		ba continue_boot
>   		 nop
> -#endif
> +
> +/* Get the machine type via the mysterious romvec node operations. */
> +get_cputype:
> +		add	%g7, 0x1c, %l1
> +		ld	[%l1], %l0
> +		ld	[%l0], %l0
> +		call	%l0
> +		 or	%g0, %g0, %o0		! next_node(0) = first_node
> +		or	%o0, %g0, %g6
> +
> +		sethi	%hi(cputypvar), %o1	! First node has cpu-arch
> +		or	%o1, %lo(cputypvar), %o1
> +		sethi	%hi(cputypval), %o2	! information, the string
> +		or	%o2, %lo(cputypval), %o2
> +		ld	[%l1], %l0		! 'compatible' tells
> +		ld	[%l0 + 0xc], %l0	! that we want 'sun4x' where
> +		call	%l0			! x is one of 'm', 'd' or 'e'.
> +		 nop				! %o2 holds pointer
> +						! to a buf where above string
> +						! will get stored by the prom.
>
>   /* Check to cputype. We may be booted on a sun4u (64 bit box),
>    * and sun4d needs special treatment.
>    */
> -
>   		set	cputypval, %o2
>   		ldub	[%o2 + 0x4], %l1
>
> @@ -467,9 +475,9 @@ sun4m_init:
>   /* This sucks, apparently this makes Vikings call prom panic, will fix later */
>   2:
>   		rd	%psr, %o1
> -		srl	%o1, 28, %o1		! Get a type of the CPU
> +		srl	%o1, PSR_IMPL_SHIFT, %o1	! Get a type of the CPU
>
> -		subcc	%o1, 4, %g0		! TI: Viking or MicroSPARC
> +		subcc	%o1, PSR_IMPL_TI, %g0		! TI: Viking or MicroSPARC
>   		be	continue_boot
>   		 nop
>

Re: [PATCH 03/15] sparc32: implement proper LEON support in head_32 (after highmem)

[Sam Ravnborg]

On Mon, May 28, 2012 at 01:52:22PM +0200, Daniel Hellstrom wrote:
> Hello Sam,
>
> I have tested the last patch series [1..15], however LEON can boot. I'm 
> trying to figure out why, this patch removes the cputypval for LEON which 
> makes the architecture detection not be executed. Since some things 
> nowdays default to LEON I suggest that the cputypval string defaults to 
> "LEON" as well, or that the check is reintroduced, or that cputypval is 
> set to "LEON" when a LEON is detected. The compatible/compatability 
> property is set to LEON on LEON systems. What to you think?
That I screwed it up :-(

The cputypval is never set for leon. I was not needed in head_32.S
and I forgot later that we did not set it for leon then.
And we use it later where we today uses the default value for leon.

I will cook up a patch for this tonight.

	Sam