Re: [PATCH] fix up request buffer reference in various scsi drivers

[Jeff Garzik <jeff@garzik.org>]

James Bottomley wrote:
> On Thu, 2006-06-08 at 09:26 -0400, Jeff Garzik wrote:
>> James Bottomley wrote:
>>> On Thu, 2006-06-08 at 07:37 -0400, Jeff Garzik wrote:
>>>> False statement, for libata.
>>> Could you amplify this statement, please ... I looked through the head
>>> of libata-dev and it seems that these are still the only two bufflen
>>> uses, which still do look obviously wrong ... I don't see where the
>>> problem in libata with this is?
>> Christoph's false statement is:  "Using the buffer and bufflen fields 
>> means they do very broken things in error handling."
>>
>> libata does not do "very broken things in error handling."
> 
> OK ... let me explain what we're doing a bit.  One of the things that
> has been going on for quite a while is that Intel (Ken Chen) has
> analyses showing that allocation clearing and freeing of the scsi_cmnd
> structure is a drag on the fast path.  This is part of the drive to slim
> it down (i.e. get rid of a lot of the fields that are only used in eh).

OK.

Since libata does EH differently than all other SCSI drivers, it 
certainly makes sense to CC the appropriate maintainers.


>>>> Please CC the author (me) or linux-ide, at least, when you touch
>>>> libata.
>>> The tradition is that for infrastructure changes like this, particularly
>>> ones that are trivial, as this appears to be, you simply copy the scsi
>>> list and don't have to sweep up the individual maintainers of each
>>> driver.
>> This is apparently an EH-related patch, and libata uses SCSI EH very 
>> differently from all other drivers.  In this case, Christoph's "very 
>> broken" justification is clearly inaccurate, and thus a CC is obviously 
>> warranted.
>>
>> Also, I remind people constantly of this, so I'm sure you and Christoph 
>> have heard the request before.
>>
>> 	Jeff, the one who forwards to linux-scsi when others forget
> 
> I appreciate that for changes you want to make to SCSI.  However, this
> particular patch touches seventeen separately maintained pieces of code
> within the SCSI subsystem, only one of which is libata.  I'm really
> reluctant to change the policy in this regard ... it's a policy we
> copied from lkml ... that for sweeping infrastructure changes we only
> send it to the list and don't have to work out who all the maintainers
> are.

If the submittor is under the impression that libata's error handling is 
"very broken", I would appreciate a clarification.  Otherwise, one must 
assume that the submittor should have CC'd linux-ide and relevant 
maintainers, because they do not understand the code they are patching.

What _precisely_ is broken, given that libata does all its own error 
handling, and ignores scsi_unjam_host() ?

	Jeff