Re: + fat-add-nobarrier-to-workaround-the-strange-behavior-of-device.patch added to -mm tree

[OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>]

Andrew Morton <akpm@linux-foundation.org> writes:

> On Wed, 3 Jul 2019 00:50:53 +0200 Christoph Hellwig <hch@lst.de> wrote:
>
>> I still very fundamentally disagree with this patch.  We did a concerted
>> effort around the other file systems to move to the device level tweak
>> and remove the badly misnamed option, so we should not add it now for
>> fat.
>
> OK, thanks, I'll leave it on hold for now.

I've checked v5.1, you mentioned "cache_type" (scsi specific
unfortunately). It calls blk_queue_write_cache(), so it calls
wbt_set_write_cache() too. So it affects to writeback throttling.

Is this intended one? At least, it was not what I expected. It is why I
asked side effect of it repeatedly in that thread.

True candidate looks like block layer's "write_cache" introduced after
v4.6. It doesn't call wbt_set_write_cache(). However, I worry if this is
intent or bug, behave like "cache_type". Did this get consensus to
replace "nobarrier" (I hope so)?

If not, I even can't recommend "write_cache", instead of "nobarrier" as
documenting, and we need real replacement to just disable flush command.

Thanks.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>