Re: [RFC] mips: Add MXU context switching support

From: PrasannaKumar Muralidharan
Date: Tue Jul 05 2016 - 06:07:27 EST


On 5 July 2016 at 04:00, Maciej W. Rozycki <macro@xxxxxxxxxx> wrote:
> On Sat, 25 Jun 2016, PrasannaKumar Muralidharan wrote:
>
>> diff --git a/arch/mips/include/asm/mxu.h b/arch/mips/include/asm/mxu.h
>> new file mode 100644
>> index 0000000..cf77cbd
>> --- /dev/null
>> +++ b/arch/mips/include/asm/mxu.h
>> @@ -0,0 +1,157 @@
>> +/*
>> + * Copyright (C) Ingenic Semiconductor
>> + * File taken from Ingenic Semiconductor's linux repo available in github
>> + *
>> + * This program is free software; you can redistribute it and/or modify it
>> + * under the terms of the GNU General Public License as published by the
>> + * Free Software Foundation; either version 2 of the License, or (at your
>> + * option) any later version.
>> + */
>> +#ifndef _ASM_MXU_H
>> +#define _ASM_MXU_H
>> +
>> +#include <asm/cpu.h>
>> +#include <asm/cpu-features.h>
>> +#include <asm/hazards.h>
>> +#include <asm/mipsregs.h>
>> +
>> +static inline void __enable_mxu(void)
>> +{
>> + unsigned int register val asm("t0");
>> + val = 3;
>> + asm volatile(".word 0x7008042f\n\t"::"r"(val));
>
> Can you please document your manually generated machine code, i.e. what
> instruction 0x7008042f actually is?

I have taken this header from vendor kernel. This instruction saves 3
to xr16 register. I will document them in the next revision.

> Also our convention has been to separate asm operands with spaces, and
> there's no need for a new line or a tab character at the end of an
> inline as GCC will add these automatically as needed, i.e.:
>
> asm volatile(".word 0x7008042f" : : "r" (val));
>
> Likewise throughout.

Will follow the convention.

>> +static inline void __save_mxu(void *tsk_void)
>> +{
>> + struct task_struct *tsk = tsk_void;
>> +
>> + register unsigned int reg_val asm("t0");
>> +
>> + asm volatile(".word 0x7008042e\n\t");
>> + tsk->thread.mxu.xr[0] = reg_val;
>> + asm volatile(".word 0x7008006e\n\t");
>> + tsk->thread.mxu.xr[1] = reg_val;
>> + asm volatile(".word 0x700800ae\n\t");
>> + tsk->thread.mxu.xr[2] = reg_val;
>> + asm volatile(".word 0x700800ee\n\t");
>> + tsk->thread.mxu.xr[3] = reg_val;
>> + asm volatile(".word 0x7008012e\n\t");
>> + tsk->thread.mxu.xr[4] = reg_val;
>> + asm volatile(".word 0x7008016e\n\t");
>> + tsk->thread.mxu.xr[5] = reg_val;
>> + asm volatile(".word 0x700801ae\n\t");
>> + tsk->thread.mxu.xr[6] = reg_val;
>> + asm volatile(".word 0x700801ee\n\t");
>> + tsk->thread.mxu.xr[7] = reg_val;
>> + asm volatile(".word 0x7008022e\n\t");
>> + tsk->thread.mxu.xr[8] = reg_val;
>> + asm volatile(".word 0x7008026e\n\t");
>> + tsk->thread.mxu.xr[9] = reg_val;
>> + asm volatile(".word 0x700802ae\n\t");
>> + tsk->thread.mxu.xr[10] = reg_val;
>> + asm volatile(".word 0x700802ee\n\t");
>> + tsk->thread.mxu.xr[11] = reg_val;
>> + asm volatile(".word 0x7008032e\n\t");
>> + tsk->thread.mxu.xr[12] = reg_val;
>> + asm volatile(".word 0x7008036e\n\t");
>> + tsk->thread.mxu.xr[13] = reg_val;
>> + asm volatile(".word 0x700803ae\n\t");
>> + tsk->thread.mxu.xr[14] = reg_val;
>> + asm volatile(".word 0x700803ee\n\t");
>> + tsk->thread.mxu.xr[15] = reg_val;
>> +}
>
> Not using an output operand with asms here?

I think the instruction saves the xr* register value to reg_val
without need for output operand.

Thanks for your review, appreciate it.