Protection information and power loss/write failures
gerry.houlder at seagate.com
Mon Mar 3 09:34:25 PST 2014
Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r1403031_f.htm">HTML-formatted message</a>
Power loss can cause lots of issues, including have a logical block contain
data that is part old data and part new data. If the last part of the
logical block has the "old data", it may also have "old protection
information", so a subsequent read attempt of the new data plus old data
plus old protection information would result in a protection check error.
Most real implementations of storage devices attempt to prevent having a
single logical block have mixed data like this but they are not always
successful. The intent of the standard is that a host should assume any
outstanding write command that did not complete before a power loss occurs
should be retried in its entirety after power is restored.
On Fri, Feb 28, 2014 at 7:20 PM, Vladislav Bolkhovitin <
vbolkhovitin at fusionio.com> wrote:
> * From the T10 Reflector (t10 at t10.org), posted by:
> * Vladislav Bolkhovitin <vbolkhovitin at fusionio.com>
> In SPC-4 "Write and unmap failures" section says that on power loss for
> those write
> commands, which have not completed when the power loss occurs, data in
> their blocks are
> indeterminate, i.e. can be "old data, new data, vendor-specific data, or
> combination thereof".
> Does it mean that protection information tags after a power loss for such
> blocks can
> mismatch their data, so subsequent reads from them can legally return tags
> mismatch errors?
> I'm asking, because it is generally assumed that writing a block on an SBC
> device is
> atomic, so the question is if this atomicity assumption applies to the PI
> tags as well?
> Hopefully, not.
> The standards are quite ambiguous in this regard, unless I'm missing
> This e-mail (and any attachments) is confidential and may be privileged.
> Any unauthorized use, copying, disclosure or dissemination of this
> communication is prohibited. If you are not the intended recipient,
> please notify the sender immediately and delete all copies of the message
> and its attachments.
> * For T10 Reflector information, send a message with
> * 'info t10' (no quotes) in the message body to majordomo at t10.org
More information about the T10