Re: [PATCH 15/18] lightnvm: pblk: fix in case of lack of lines

[Igor Konopko <igor.j.konopko@intel.com>]

On 18.03.2019 08:42, Javier González wrote:
>> On 14 Mar 2019, at 17.04, Igor Konopko <igor.j.konopko@intel.com> wrote:
>>
>> In case when mapping fails (called from writer thread) due to lack of
>> lines, currently we are calling pblk_pipeline_stop(), which waits
>> for pending write IOs, so it will lead to the deadlock. Switching
>> to __pblk_pipeline_stop() in that case instead will fix that.
>>
>> Signed-off-by: Igor Konopko <igor.j.konopko@intel.com>
>> ---
>> drivers/lightnvm/pblk-map.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/lightnvm/pblk-map.c b/drivers/lightnvm/pblk-map.c
>> index 5408e32..afc10306 100644
>> --- a/drivers/lightnvm/pblk-map.c
>> +++ b/drivers/lightnvm/pblk-map.c
>> @@ -46,7 +46,7 @@ static int pblk_map_page_data(struct pblk *pblk, unsigned int sentry,
>> 		pblk_line_close_meta(pblk, prev_line);
>>
>> 		if (!line) {
>> -			pblk_pipeline_stop(pblk);
>> +			__pblk_pipeline_stop(pblk);
>> 			return -ENOSPC;
>> 		}
>>
>> --
>> 2.9.5
> 
> Have you seeing this problem?
> 
> Before checking if there is a line, we are closing metadata for the
> previous line, so all inflight I/Os should be clear. Can you develop on
> the case in which this would happen?

So we have following sequence: pblk_pipeline_stop() -> 
__pblk_pipeline_flush() -> pblk_flush_writer() -> wait for emptying 
round buffer.
This will never complete, since we still have some RB entries, which 
cannot be written since writer thread is blocked with waiting inside 
pblk_flush_writer().

Am I missing sth?

>