Re: [RFC 4/4] ARC: uaccess: use optimized generic __strnlen_user/__strncpy_from_user

[Vineet Gupta]

On 1/14/20 12:42 PM, Arnd Bergmann wrote:
> On Tue, Jan 14, 2020 at 9:08 PM Vineet Gupta <Vineet.Gupta1@xxxxxxxxxxxx> wrote:
> 
>> diff --git a/arch/arc/include/asm/word-at-a-time.h b/arch/arc/include/asm/word-at-a-time.h
>> new file mode 100644
>> index 000000000000..00e92be70987
>> --- /dev/null
>> +++ b/arch/arc/include/asm/word-at-a-time.h
>> @@ -0,0 +1,49 @@
>> +/* SPDX-License-Identifier: GPL-2.0-only */
>> +/*
>> + * Copyright (C) 2020 Synopsys Inc.
>> + */
>> +#ifndef __ASM_ARC_WORD_AT_A_TIME_H
>> +#define __ASM_ARC_WORD_AT_A_TIME_H
>> +
>> +#ifdef __LITTLE_ENDIAN__
>> +
>> +#include <linux/kernel.h>
>> +
>> +struct word_at_a_time {
>> +       const unsigned long one_bits, high_bits;
>> +};
> 
> What's wrong with the generic version on little-endian? Any
> chance you can find a way to make it work as well for you as
> this copy?

find_zero() by default doesn't use pop count instructions. I didn't like the copy
either but wasn't sure of the best way to make this 4 API interface reusable. Are
you suggesting we allow partial over-ride starting with #ifndef find_zero ?

>> +static inline unsigned long find_zero(unsigned long mask)
>> +{
>> +#ifdef CONFIG_64BIT
>> +       return fls64(mask) >> 3;
>> +#else
>> +       return fls(mask) >> 3;
>> +#endif
> 
> The CONFIG_64BIT check not be needed, unless you are adding
> support for 64-bit ARC really soon.

:-) Indeed that was the premise !

Thx for the quick review.

-Vineet