[PATCH] MIPS: DEC: Fix an int-handler.S CPU_DADDI_WORKAROUNDS regression

[PATCH] MIPS: DEC: Fix an int-handler.S CPU_DADDI_WORKAROUNDS regression

[Maciej W. Rozycki]

Fix a commit 3021773c7c3e ("MIPS: DEC: Avoid la pseudo-instruction in 
delay slots") regression and remove assembly errors:

arch/mips/dec/int-handler.S: Assembler messages:
arch/mips/dec/int-handler.S:162: Error: Macro used $at after ".set noat"
arch/mips/dec/int-handler.S:163: Error: Macro used $at after ".set noat"
arch/mips/dec/int-handler.S:229: Error: Macro used $at after ".set noat"
arch/mips/dec/int-handler.S:230: Error: Macro used $at after ".set noat"

triggering with with the CPU_DADDI_WORKAROUNDS option set and the DADDIU 
instruction.  This is because with that option in place the instruction 
becomes a macro, which expands to an LI/DADDU (or actually ADDIU/DADDU) 
sequence that uses $at as a temporary register.  Use ADDIU instead then, 
which is equivalent when used with LUI to compose the intermediate 
32-bit values.

Cc: stable@vger.kernel.org # 4.8+
Fixes: 3021773c7c3e ("MIPS: DEC: Avoid la pseudo-instruction in delay slots")
Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
---
 arch/mips/dec/int-handler.S |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

linux-mips-dec-int-handler-dla.diff
Index: linux-sfr-4maxp64/arch/mips/dec/int-handler.S
===================================================================
--- linux-sfr-4maxp64.orig/arch/mips/dec/int-handler.S
+++ linux-sfr-4maxp64/arch/mips/dec/int-handler.S
@@ -159,8 +159,8 @@
 		.set	noat
 		lui	t1, %highest(cpu_mask_nr_tbl)
 		lui	AT, %hi(cpu_mask_nr_tbl)
-		daddiu	t1, t1, %higher(cpu_mask_nr_tbl)
-		daddiu	AT, AT, %lo(cpu_mask_nr_tbl)
+		addiu	t1, t1, %higher(cpu_mask_nr_tbl)
+		addiu	AT, AT, %lo(cpu_mask_nr_tbl)
 		dsll	t1, 32
 		daddu	t1, t1, AT
 		.set	pop
@@ -226,8 +226,8 @@
 		.set	noat
 		lui	t1, %highest(asic_mask_nr_tbl)
 		lui	AT, %hi(asic_mask_nr_tbl)
-		daddiu	t1, t1, %higher(asic_mask_nr_tbl)
-		daddiu	AT, AT, %lo(asic_mask_nr_tbl)
+		addiu	t1, t1, %higher(asic_mask_nr_tbl)
+		addiu	AT, AT, %lo(asic_mask_nr_tbl)
 		dsll	t1, 32
 		daddu	t1, t1, AT
 		.set	pop

Re: [PATCH] MIPS: DEC: Fix an int-handler.S CPU_DADDI_WORKAROUNDS regression

[Maciej W. Rozycki]

On Sun, 9 Jul 2017, Maciej W. Rozycki wrote:

> Fix a commit 3021773c7c3e ("MIPS: DEC: Avoid la pseudo-instruction in 
> delay slots") regression and remove assembly errors:
> 
> arch/mips/dec/int-handler.S: Assembler messages:
> arch/mips/dec/int-handler.S:162: Error: Macro used $at after ".set noat"
> arch/mips/dec/int-handler.S:163: Error: Macro used $at after ".set noat"
> arch/mips/dec/int-handler.S:229: Error: Macro used $at after ".set noat"
> arch/mips/dec/int-handler.S:230: Error: Macro used $at after ".set noat"
> 
> triggering with with the CPU_DADDI_WORKAROUNDS option set and the DADDIU 
> instruction.  This is because with that option in place the instruction 
> becomes a macro, which expands to an LI/DADDU (or actually ADDIU/DADDU) 
> sequence that uses $at as a temporary register.  Use ADDIU instead then, 
> which is equivalent when used with LUI to compose the intermediate 
> 32-bit values.

 64-bit DEC configurations unconditionally select CPU_DADDI_WORKAROUNDS, 
which in turn requires KBUILD_SYM32 or a build error is triggered in our 
top-level arch Makefile.  Therefore this issue can be handled better, by 
checking for KBUILD_64BIT_SYM32 and then the problematic 64-bit leg of 
the #ifdef can be reduced to #error, as not supposed to be ever enabled.

 Withdrawing this patch then, I'll send a different one soon.

  Maciej