[Devel] [RFC PATCH vz9 v6 22/62] dm-ploop: speed up ploop_prepare_bat_update()

[Alexander Atanasov]

On 20.01.25 6:50, Pavel Tikhomirov wrote:
> 
> 
> On 12/6/24 05:55, Alexander Atanasov wrote:
>> From: Andrey Zhadchenko <andrey.zhadchenko at virtuozzo.com>
>>
>> Drop ploop_cluster_is_in_top_delta(), because
>>   - it internally searches md on every call
>>   - it takes read lock every time
>>
>> https://virtuozzo.atlassian.net/browse/VSTOR-91817
>> Signed-off-by: Andrey Zhadchenko <andrey.zhadchenko at virtuozzo.com>
>> ---
>>   drivers/md/dm-ploop-map.c | 15 +++++++++++----
>>   1 file changed, 11 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/md/dm-ploop-map.c b/drivers/md/dm-ploop-map.c
>> index c89e28aa18f9..de885049e3f9 100644
>> --- a/drivers/md/dm-ploop-map.c
>> +++ b/drivers/md/dm-ploop-map.c
>> @@ -899,13 +899,14 @@ static void ploop_bat_write_complete(struct pio 
>> *pio, void *piwb_ptr,
>>   static int ploop_prepare_bat_update(struct ploop *ploop, struct 
>> md_page *md,
>>                       enum piwb_type type)
>>   {
>> -    u32 i, off, last, *bat_entries;
>> +    u32 i, off, last, *bat_entries, clu;
>>       struct ploop_index_wb *piwb;
>>       bool is_last_page = true;
>>       u32 page_id = md->id;
>>       struct page *page;
>>       struct pio *pio;
>>       map_index_t *to;
>> +    u8 level;
>>       piwb = kmalloc(sizeof(*piwb), GFP_NOIO);
>>       if (!piwb)
>> @@ -928,6 +929,8 @@ static int ploop_prepare_bat_update(struct ploop 
>> *ploop, struct md_page *md,
>>       piwb->page_id = page_id;
>>       to = piwb->kmpage;
>> +
>> +    read_lock_irq(&md->lock);
>>       memcpy((void *)to, bat_entries, PAGE_SIZE);
>>       /* Absolute number of first index in page (negative for page#0) */
>> @@ -946,10 +949,14 @@ static int ploop_prepare_bat_update(struct ploop 
>> *ploop, struct md_page *md,
>>       /* Copy BAT (BAT goes right after hdr, see .ctr) */
>>       for (; i < last; i++) {
>> -        if (ploop_cluster_is_in_top_delta(ploop, i + off))
>> -            continue;
>> -        to[i] = 0;
>> +        level = md->bat_levels[i];
> 
> Why do we assume that md page will not change when iterating from 
> PLOOP_MAP_OFFSET and until (ploop->nr_bat_entries - (page_id * PAGE_SIZE 
> / sizeof(map_index_t) - PLOOP_MAP_OFFSET) ? If it's true, we need 
> explanation in commit message. If false, we need to update md each time 
> as ploop_cluster_is_in_top_delta did.

this can change on grow or shrink - but when such operation is in 
progress, there are no pios in flight and pio submission is suspended - 
so it can not change. Change is applied at the end of operation.



> 
>> +        clu = READ_ONCE(bat_entries[i]);
>> +
>> +        if (clu == BAT_ENTRY_NONE || level < ploop_top_level(ploop))
>> +            to[i] = 0;
>>       }
>> +    read_unlock_irq(&md->lock);
>> +
>>       if (is_last_page) {
>>       /* Fill tail of page with 0 */
>>           for (i = last; i < PAGE_SIZE / sizeof(map_index_t); i++)
> 

-- 
Regards,
Alexander Atanasov