Am 10.07.2013 16:58, schrieb Kevin Wolf: > Am 10.07.2013 um 16:49 hat Peter Lieven geschrieben: >> Am 10.07.2013 16:28, schrieb Kevin Wolf: >>> Am 10.07.2013 um 16:04 hat Peter Lieven geschrieben: >>>> Am 10.07.2013 13:33, schrieb Kevin Wolf: >>>>> Am 27.06.2013 um 15:11 hat Peter Lieven geschrieben: >>>>>> if the target does not support UNMAP or the request >>>>>> is too big silently ignore the discard request. >>>>>> >>>>>> Signed-off-by: Peter Lieven <p...@kamp.de> >>>>> Why not loop for the "too big" case? You can probably use the same logic >>>>> for unmapping the whole device in .bdrv_create and here. >>>> right, but looping in an aio function seemed not so trivial to me. >>>> it seems more and more obvious to me that the best would be to change >>>> all the remaining aio routines to co routines. >>> The pattern for AIO functions is that the real work of submitting >>> requests is done in the AIO callback, and it submits new AIO requests >>> calling back into the same callback as long as acb->remaining_secs > 0 >>> (or something like that). >>> >>> You can still see that kind of thing alive in qed_aio_next_io(), (most >>> of?) the rest is converted to coroutines because it makes the code look >>> nicer. >> would you agree if I leave the easy version in just to fix the potential >> problems if iscsi_aio_discard is called with too high nb_sectors or >> on a storage where UNMAP is unsupported. >> >> I will add a TODO with the comment that the limit of iscsi->max_unmap should >> be replaced by a loop once the routine is replaced by a coroutine? > Meh, another pony I don't get... ;-) > > Leaving a TODO comment for now is okay with me. forget about the TODO. Looking at what I have already got, implementing iscsi_co_discard is low hanging fruit. I will implement it directly. With the too big loop and all the benefits of the co routines.
Peter
