Messages in this thread | | | From | Jeff Moyer <> | Subject | Re: [PATCH] block: Force sector and nr_sects to device alignment and granularity. | Date | Wed, 12 Mar 2014 15:33:58 -0400 |
| |
Frank Mayhar <fmayhar@google.com> writes:
> On Wed, 2014-03-12 at 14:20 -0400, Jeff Moyer wrote: >> but you managed to read my mind well enough. The question is how high >> up the stack do you put the logic for this? Is it worth it to duplicate >> the checks in the OS that are already done on the device? I don't >> know. Martin, do you have an opinion on this? > > Well, my opinion (and I suspect that Ted agrees with me to at least some > extent) is that this is where it should be, i.e. in the block layer, in > the place that already knows about and deals with alignment and > granularity. Sure, you could leave it to the device itself but it seems > reasonable to take care of this here for two reasons: First, doing this > means that if a TRIM is issued it will be successful and the intent of
No, TRIM is advisory, even for well-formed TRIMs. I guess you could alter the definition of successful and have a correct statement there.
> the discard will be at least partly satisfied. Second, we're already > doing most of the computations and making decisions based on the > alignment and granularity anyway, so the overhead is pretty negligible > (and if the discard size goes to zero we short-circuit the process and > never go to the device at all).
Sure, there's no sense getting information from the device and not using it. I guess you've talked me into it.
Cheers, Jeff
| |