Clarification - export of scsi_finish_command and scsi_times_out

Clarification - export of scsi_finish_command and scsi_times_out

[Mike Anderson]

On a previous thread to resolve a libata issue scsi_finish_command was
exported. The thread can be reference at the url below.

http://marc.theaimsgroup.com/?t=107799572900002&r=1&w=2

My question is why did we export scsi_finish_command and not a wrapper
that included scsi_decide_disposition. If a LLDD did not have a
eh_strategy routine it is possible post recovery the io would be
retried. Is it because it was believed that these timed out commands
handled by the libata eh_strategy_handler would most likely not succeed
on retry?

Also there was a previous patch posted by Christoph to allow a LLDD to
handle timeouts directly instead of waiting for the host_failed count to
be reached where they could then be handled in a eh_strategy_handler.
http://marc.theaimsgroup.com/?l=linux-scsi&m=107460498631200&w=2

What is the current thought on this patch.

-andmike
--
Michael Anderson
andmike@us.ibm.com

Re: Clarification - export of scsi_finish_command and scsi_times_out

[James Bottomley]

On Tue, 2004-06-08 at 19:18, Mike Anderson wrote:
> On a previous thread to resolve a libata issue scsi_finish_command was
> exported. The thread can be reference at the url below.
> 
> http://marc.theaimsgroup.com/?t=107799572900002&r=1&w=2
> 
> My question is why did we export scsi_finish_command and not a wrapper
> that included scsi_decide_disposition. If a LLDD did not have a
> eh_strategy routine it is possible post recovery the io would be
> retried. Is it because it was believed that these timed out commands
> handled by the libata eh_strategy_handler would most likely not succeed
> on retry?

This was just a quick hack to get libata working.  The particular
problem had no need of decide_disposition, it just wanted an error
return if there was a problem.  We probably would need to export decide
disposition for a proper user of eh_strategy_handler.  On the other
hand, I was wondering if we couldn't deprecate eh_strategy_handler and
go with error handling within transport classes instead as a future
direction.

> Also there was a previous patch posted by Christoph to allow a LLDD to
> handle timeouts directly instead of waiting for the host_failed count to
> be reached where they could then be handled in a eh_strategy_handler.
> http://marc.theaimsgroup.com/?l=linux-scsi&m=107460498631200&w=2
> 
> What is the current thought on this patch.

Looks fine as a proposal.  However, it's only a first draft patch and we
have no putative users to hone it into a finished API.

James