[PATCH 12/16] zonefs: allow non power of 2 zoned devices

[Damien Le Moal]

On 4/29/22 00:54, Pankaj Raghav wrote:
> On 2022-04-28 01:39, Damien Le Moal wrote:
>> On 4/28/22 01:02, Pankaj Raghav wrote:
>>> The zone size shift variable is useful only if the zone sizes are known
>>> to be power of 2. Remove that variable and use generic helpers from
>>> block layer to calculate zone index in zonefs
>>
>> Period missing at the end of the sentence.
>>
> Ack
>> What about zonefs-tools and its test suite ? Is everything still OK on
>> that front ? I suspect not...
>>
> I don't know why you assume that :). Zonefs had only one place that had
> the assumption of po2 zsze sectors:
> if (nr_zones < dev.nr_zones) {
> 	size_t runt_sectors = dev.capacity & (dev.zone_nr_sectors - 1);
> 
> In my local tree I changed it and I was able to run zonefs tests for non
> po2 zone device. I have also mentioned it in my cover letter:
> ```
> ZoneFS:
> zonefs-tests.sh from zonefs-tools were run with no failures.
> ```

This is still not convincing given the code I saw. Additional test cases
need to be added with data verification & concurrent regular writes also
sent while doing copy to verify locking.

Which also reminds me that I have not seen any change to mq-deadline zone
write locking for this series. What is the assumption ? That users should
not be issuing writes when a copy is on-going ? What a bout the reverse
case ? at the very least, it seems that blk_issue_copy() should be taking
the zone write lock.

> I will make sure to add my private tree for zonefs in my cover letter in
> the next rev. But even without that change, a typical emulated npo2
> device should work fine because the changes are applicable only for
> "runt" zones.


-- 
Damien Le Moal
Western Digital Research