[PATCH 01/16] block: make blkdev_nr_zones and blk_queue_zone_no generic for npo2 zsze

Pankaj Raghav p.raghav at samsung.com
Wed May 4 01:35:04 PDT 2022


On 2022-05-03 18:37, Bart Van Assche wrote:
>>       sector_t zone_sectors = blk_queue_zone_sectors(disk->queue);
>> +    sector_t capacity = get_capacity(disk);
>>         if (!blk_queue_is_zoned(disk->queue))
>>           return 0;
>> -    return (get_capacity(disk) + zone_sectors - 1) >>
>> ilog2(zone_sectors);
>> +
>> +    if (is_power_of_2(zone_sectors))
>> +        return (capacity + zone_sectors - 1) >>
>> +               ilog2(zone_sectors);
>> +
>> +    return div64_u64(capacity + zone_sectors - 1, zone_sectors);
>>   }
>>   EXPORT_SYMBOL_GPL(blkdev_nr_zones);
> 
> Does anyone need support for more than 4 billion sectors per zone? If
> not, do_div() should be sufficient.
> 
You are absolutely right but blk_queue_zone_sectors explicitly changed
the return type to uint32_t to sector_t in this commit:
'113ab72  block: Fix potential overflow in blk_report_zones()'.

I initially did have a do_div but later changed to div64_u64 to avoid
any implicit down casting even though zone_sectors will be always
limited to an unsigned int.
>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
>> index 60d016138997..c4e4c7071b7b 100644
>> --- a/include/linux/blkdev.h
>> +++ b/include/linux/blkdev.h
>> @@ -665,9 +665,15 @@ static inline unsigned int
>> blk_queue_nr_zones(struct request_queue *q)
>>   static inline unsigned int blk_queue_zone_no(struct request_queue *q,
>>                            sector_t sector)
>>   {
>> +    sector_t zone_sectors = blk_queue_zone_sectors(q);
>> +
>>       if (!blk_queue_is_zoned(q))
>>           return 0;
>> -    return sector >> ilog2(q->limits.chunk_sectors);
>> +
>> +    if (is_power_of_2(zone_sectors))
>> +        return sector >> ilog2(zone_sectors);
>> +
>> +    return div64_u64(sector, zone_sectors);
>>   }
> 
> Same comment here.
> 
> Thanks,
> 
> Bart.



More information about the Linux-nvme mailing list