Re: [PATCH 1/2] block: RCU protect disk->conv_zones_bitmap

[Damien Le Moal <dlemoal@kernel.org>]

On 11/7/24 09:02, Jens Axboe wrote:
> On 11/6/24 4:44 PM, Damien Le Moal wrote:
>> On 11/7/24 08:20, Bart Van Assche wrote:
>>> On 11/6/24 3:13 PM, Damien Le Moal wrote:
>>>> diff --git a/block/blk-zoned.c b/block/blk-zoned.c
>>>> index a287577d1ad6..7a7855555d6d 100644
>>>> --- a/block/blk-zoned.c
>>>> +++ b/block/blk-zoned.c
>>>> @@ -350,9 +350,14 @@ int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
>>>>   
>>>>   static inline bool disk_zone_is_conv(struct gendisk *disk, sector_t sector)
>>>>   {
>>>> -	if (!disk->conv_zones_bitmap)
>>>> -		return false;
>>>> -	return test_bit(disk_zone_no(disk, sector), disk->conv_zones_bitmap);
>>>> +	bool is_conv;
>>>> +
>>>> +	rcu_read_lock();
>>>> +	is_conv = disk->conv_zones_bitmap &&
>>>> +		test_bit(disk_zone_no(disk, sector), disk->conv_zones_bitmap);
>>>> +	rcu_read_unlock();
>>>> +
>>>> +	return is_conv;
>>>>   }
>>>
>>> The above code can be simplified significantly by using guard(rcu).
>>
>> I personally dislike very much annotations that hide code. So unless
>> Jens prefers using guard(rcu), I would prefer leaving the code as it
>> is.
> 
> I don't mind it, and I do use it myself when it makes sense - but imho
> it's up to the person writing the code, particularly when it's their
> code in the first place.

OK, then I would prefer leaving the code as is :)

-- 
Damien Le Moal
Western Digital Research