Hello,
On 03/02/2010 11:44 AM, Daniel Taylor wrote:
> If I can provide some test hardware, I'll need an NDA.
>
> Is there someone who has the appropriate lab and can sign one?
Richardson has been providing test drives to me and I've forwarded
some to Arvin Schnell and now that I have more 4k (512 logical) drives
I can forward more to any Novell people. If anyone wants some, please
let me know.
> FYI, here's the summary I have been providing (some of which may, finally,
> be obsolete):
>
> ...
> Three flavors of drives with 4096-byte sectors: 4096 physical/logical; 4096
> physical/512 logical, LBA offset 0 within physical; 4096 physical/512
> logical, LBA offset 1 within the physical. Around here they are often
> referred to as AF-4K, AF-0, and AF-1, respectively. This is complicated by
> two facters, drive capacities larger that 2TiB, and MBR vs GPT.
Are you guys planning on shipping 4k logical ATA drives? Seriously?
> The problem is not usually the kernel (although I have one case that I'm
> debugging now which seems to be a kernel issue).
>
> The partitioning/installation tools do not work well with two of the three
> flavors. Currently, parted explicity complains about AF-4K, and really does
> not work. Sfdisk cannot handle AF-0 and AF-1 for drives larger than 2TiB.
>
> The default installers do not check whether a drive is AF-0 or AF-1, and
> tend to use obsolete BIOS alignments (track-aligned for a track with 63
> sectors/track), which work well enough for AF-1, but utterly destroy
> performance with AF-0, since the drives are continually performing
> read-modify-write on two physical sectors for each file system block write.
> The problem is that the file system blocks are mis-aligned with regard to
> the drive physical sectors.
> ...
>
> One more new thing, there also appear to be drives that actually have 4K
> sectors, but report 512 physical and logical. Those appear to be AF-0. It
> might be a good idea for the installers to assume 4K sectors on all
> late-model drives.
If it's just a glitch on certain early models, easier workaround might
be blacklisting those drives in the driver so that the kernel can
report the correct physical sector size and alignment via sysfs to
userland.
Thanks.
--
tejun
--
To unsubscribe, e-mail: opensuse-kernel+unsubscribe(a)opensuse.org
For additional commands, e-mail: opensuse-kernel+help(a)opensuse.org
