Messages in this thread | | | Date | Sun, 12 Jan 2014 13:49:36 +0200 | From | Sagi Grimberg <> | Subject | Re: [PATCH 07/14] target/sbc: Add P_TYPE + PROT_EN bits to READ_CAPACITY_16 |
| |
On 1/10/2014 10:46 PM, Martin K. Petersen wrote: >>>>>> "Andy" == Andy Grover <agrover@redhat.com> writes: > Andy> Yes, don't you need FORMAT UNIT because protection information is > Andy> going to mean the pi-enabled lun will need to report less blocks? > > Modern disk drives won't shrink when you reformat them with PI. This is > a result of an IDEMA agreement about LBA counts. > > And if you create a 10GB PI LUN on an array you'll get 10GB for data. > > Andy> The ramdisk backstore changes in this series allocate extra space > Andy> for PI info, but my understanding was that especially for > Andy> emulation with block and fileio backstores, everything needs to go > Andy> in the same amount of space. > > For both file and block I'd recommend we store the PI in a separate > block device or file unless the backing device is PI-capable. > > Andy> Furthermore, if we want PI info stored along with the blocks, then > Andy> block and fileio backstore formats are no longer going to be 1:1 > Andy> -- requiring offset calculations, non-aligned read-modify-write, > Andy> and all that unpleasantness to be handled? > > I only think interleaved makes sense if you're passing the PI through > instead of emulating. >
I agree, I implemented interleaved mode just as a proof of concept that our HW can perform offload in that manner. I assume we can stick with non-interleaved, although it can be added as a user option.
Sagi.
| |