On 05/27, Christoph Hellwig wrote: > On Fri, May 27, 2022 at 06:06:08PM -0700, Jaegeuk Kim wrote: > > I think there's still some benefit to use FUA such as small chunk writes > > for checkpoint. > > Did you measure if there is? Because some SSDs basically implemented > FUA as an implied flush after the write, in which case it would not > really help there either (but also not hurt). > > But as the previous two maintainers already said - this needs quirking > at the driver layer, not in the submitter. Thanks, I indeed measured this using UFS, and it turned out cache_flush is better than FUA all the time like this. Hence, I posted a quirk [1]. Write(us/KB) 4 64 256 1024 2048 FUA 873.792 754.604 995.624 1011.67 1067.99 CACHE_FLUSH 824.703 712.98 800.307 1019.5 1037.37 [1] https://lore.kernel.org/linux-scsi/20220531201053.3300018-1-jaegeuk@xxxxxxxxxx/T/#u
